## Introduction

This is the post of 2020, so happy new year to you all !

I’m a huge fan of LLVM since 11 years ago when I started playing with it to JIT data structures such as AVLs, then later to JIT restricted AST trees and to JIT native code from TensorFlow graphs. Since then, LLVM evolved into one of the most important compiler framework ecosystem and is used nowadays by a lot of important open-source projects.

One cool project that I recently became aware of is Gandiva. Gandiva was developed by Dremio and then later donated to Apache Arrow (kudos to Dremio team for that). The main idea of Gandiva is that it provides a compiler to generate LLVM IR that can operate on batches of Apache Arrow. Gandiva was written in C++ and comes with a lot of different functions implemented to build an expression tree that can be JIT’ed using LLVM. One nice feature of this design is that it can use LLVM to automatically optimize complex expressions, add native target platform vectorization such as AVX while operating on Arrow batches and execute native code to evaluate the expressions.

The image below gives an overview of Gandiva:

In this post I’ll build a very simple expression parser supporting a limited set of operations that I will use to filter a Pandas DataFrame.

## Building simple expression with Gandiva

In this section I’ll show how to create a simple expression manually using tree builder from Gandiva.

### Using Gandiva Python bindings to JIT and expression

Before building our parser and expression builder for expressions, let’s manually build a simple expression with Gandiva. First, we will create a simple Pandas DataFrame with numbers from 0.0 to 9.0:

import pandas as pd
import pyarrow as pa
import pyarrow.gandiva as gandiva

# Create a simple Pandas DataFrame
df = pd.DataFrame({"x": [1.0 * i for i in range(10)]})
table = pa.Table.from_pandas(df)
schema = pa.Schema.from_pandas(df)

We converted the DataFrame to an Arrow Table, it is important to note that in this case it was a zero-copy operation, Arrow isn’t copying data from Pandas and duplicating the DataFrame. Later we get the schemafrom the table, that contains column types and other metadata.

After that, we want to use Gandiva to build the following expression to filter the data:

(x > 2.0) and (x < 6.0)

This expression will be built using nodes from Gandiva:

builder = gandiva.TreeExprBuilder()

# Reference the column "x"
node_x = builder.make_field(table.schema.field("x"))

# Make two literals: 2.0 and 6.0
two = builder.make_literal(2.0, pa.float64())
six = builder.make_literal(6.0, pa.float64())

# Create a function for "x > 2.0"
gt_five_node = builder.make_function("greater_than",
[node_x, two],
pa.bool_())

# Create a function for "x < 6.0"
lt_ten_node = builder.make_function("less_than",
[node_x, six],
pa.bool_())
# Create an "and" node, for "(x > 2.0) and (x < 6.0)"
and_node = builder.make_and([gt_five_node, lt_ten_node])

# Make the expression a condition and create a filter
condition = builder.make_condition(and_node)
filter_ = gandiva.make_filter(table.schema, condition)

This code now looks a little more complex but it is easy to understand. We are basically creating the nodes of a tree that will represent the expression we showed earlier. Here is a graphical representation of what it looks like:

### Inspecting the generated LLVM IR

Unfortunately,  haven’t found a way to dump the LLVM IR that was generated using the Arrow’s Python bindings, however, we can just use the C++ API to build the same tree and then look at the generated LLVM IR:

auto field_x = field("x", float32());
auto schema = arrow::schema({field_x});

auto node_x = TreeExprBuilder::MakeField(field_x);

auto two = TreeExprBuilder::MakeLiteral((float_t)2.0);
auto six = TreeExprBuilder::MakeLiteral((float_t)6.0);

auto gt_five_node = TreeExprBuilder::MakeFunction("greater_than",
{node_x, two}, arrow::boolean());

auto lt_ten_node = TreeExprBuilder::MakeFunction("less_than",
{node_x, six}, arrow::boolean());

auto and_node = TreeExprBuilder::MakeAnd({gt_five_node, lt_ten_node});
auto condition = TreeExprBuilder::MakeCondition(and_node);

std::shared_ptr<Filter> filter;
auto status = Filter::Make(schema, condition, TestConfiguration(), &filter);

The code above is the same as the Python code, but using the C++ Gandiva API. Now that we built the tree in C++, we can get the LLVM Module and dump the IR code for it. The generated IR is full of boilerplate code and the JIT’ed functions from the Gandiva registry, however the important parts are show below:

; Function Attrs: alwaysinline norecurse nounwind readnone ssp uwtable
define internal zeroext i1 @less_than_float32_float32(float, float) local_unnamed_addr #0 {
%3 = fcmp olt float %0, %1
ret i1 %3
}

; Function Attrs: alwaysinline norecurse nounwind readnone ssp uwtable
define internal zeroext i1 @greater_than_float32_float32(float, float) local_unnamed_addr #0 {
%3 = fcmp ogt float %0, %1
ret i1 %3
}

(...)
%x = load float, float* %11
%greater_than_float32_float32 = call i1 @greater_than_float32_float32(float %x, float 2.000000e+00)
(...)
%x11 = load float, float* %15
%less_than_float32_float32 = call i1 @less_than_float32_float32(float %x11, float 6.000000e+00)



As you can see, on the IR we can see the call to the functions less_than_float32_float_32 and greater_than_float32_float32that are the (in this case very simple) Gandiva functions to do float comparisons. Note the specialization of the function by looking at the function name prefix.

What is quite interesting is that LLVM will apply all optimizations in this code and it will generate efficient native code for the target platform while Godiva and LLVM will take care of making sure that memory alignment will be correct for extensions such as AVX to be used for vectorization.

This IR code I showed isn’t actually the one that is executed, but the optimized one. And in the optimized one we can see that LLVM inlined the functions, as shown in a part of the optimized code below:

%x.us = load float, float* %10, align 4
%11 = fcmp ogt float %x.us, 2.000000e+00
%12 = fcmp olt float %x.us, 6.000000e+00
%not.or.cond = and i1 %12, %11


You can see that the expression is now much simpler after optimization as LLVM applied its powerful optimizations and inlined a lot of Gandiva funcions.

## Building a Pandas filter expression JIT with Gandiva

Now we want to be able to implement something similar as the Pandas’ DataFrame.query()function using Gandiva. The first problem we will face is that we need to parse a string such as (x > 2.0) and (x < 6.0), later we will have to build the Gandiva expression tree using the tree builder from Gandiva and then evaluate that expression on arrow data.

Now, instead of implementing a full parsing of the expression string, I’ll use the Python AST module to parse valid Python code and build an Abstract Syntax Tree (AST) of that expression, that I’ll be later using to emit the Gandiva/LLVM nodes.

The heavy work of parsing the string will be delegated to Python AST module and our work will be mostly walking on this tree and emitting the Gandiva nodes based on that syntax tree. The code for visiting the nodes of this Python AST tree and emitting Gandiva nodes is shown below:

class LLVMGandivaVisitor(ast.NodeVisitor):
def __init__(self, df_table):
self.table = df_table
self.builder = gandiva.TreeExprBuilder()
self.columns = {f.name: self.builder.make_field(f)
for f in self.table.schema}
self.compare_ops = {
"Gt": "greater_than",
"Lt": "less_than",
}
self.bin_ops = {
"BitAnd": self.builder.make_and,
"BitOr": self.builder.make_or,
}

def visit_Module(self, node):
return self.visit(node.body[0])

def visit_BinOp(self, node):
left = self.visit(node.left)
right = self.visit(node.right)
op_name = node.op.__class__.__name__
gandiva_bin_op = self.bin_ops[op_name]
return gandiva_bin_op([left, right])

def visit_Compare(self, node):
op = node.ops[0]
op_name = op.__class__.__name__
gandiva_comp_op = self.compare_ops[op_name]
comparators = self.visit(node.comparators[0])
left = self.visit(node.left)
return self.builder.make_function(gandiva_comp_op,
[left, comparators], pa.bool_())

def visit_Num(self, node):
return self.builder.make_literal(node.n, pa.float64())

def visit_Expr(self, node):
return self.visit(node.value)

def visit_Name(self, node):
return self.columns[node.id]

def generic_visit(self, node):
return node

def evaluate_filter(self, llvm_mod):
condition = self.builder.make_condition(llvm_mod)
filter_ = gandiva.make_filter(self.table.schema, condition)
result = filter_.evaluate(self.table.to_batches()[0],
pa.default_memory_pool())
arr = result.to_array()
pd_result = arr.to_numpy()
return pd_result

@staticmethod
def gandiva_query(df, query):
df_table = pa.Table.from_pandas(df)
llvm_gandiva_visitor = LLVMGandivaVisitor(df_table)
mod_f = ast.parse(query)
llvm_mod = llvm_gandiva_visitor.visit(mod_f)
results = llvm_gandiva_visitor.evaluate_filter(llvm_mod)
return results

As you can see, the code is pretty straightforward as I’m not supporting every possible Python expressions but a minor subset of it. What we do in this class is basically a conversion of the Python AST nodes such as Comparators and BinOps (binary operations) to the Gandiva nodes. I’m also changing the semantics of the & and the | operators to represent AND and OR respectively, such as in Pandas query()function.

### Register as a Pandas extension

The next step is to create a simple Pandas extension using the gandiva_query() method that we created:

@pd.api.extensions.register_dataframe_accessor("gandiva")
class GandivaAcessor:
def __init__(self, pandas_obj):
self.pandas_obj = pandas_obj

def query(self, query):
return LLVMGandivaVisitor.gandiva_query(self.pandas_obj, query)

And that is it, now we can use this extension to do things such as:

df = pd.DataFrame({"a": [1.0 * i for i in range(nsize)]})
results = df.gandiva.query("a > 10.0")

As we have registered a Pandas extension called gandiva that is now a first-class citizen of the Pandas DataFrames.

Let’s create now a 5 million floats DataFrame and use the new query() method to filter it:

df = pd.DataFrame({"a": [1.0 * i for i in range(50000000)]})
df.gandiva.query("a < 4.0")

# This will output:
#     array([0, 1, 2, 3], dtype=uint32)

Note that the returned values are the indexes satisfying the condition we implemented, so it is different than the Pandas query()that returns the data already filtered.

I did some benchmarks and found that Gandiva is usually always faster than Pandas, however I’ll leave proper benchmarks for a next post on Gandiva as this post was to show how you can use it to JIT expressions.

That’s it ! I hope you liked the post as I enjoyed exploring Gandiva. It seems that we will probably have more and more tools coming up with Gandiva acceleration, specially for SQL parsing/projection/JITing. Gandiva is much more than what I just showed, but you can get started now to understand more of its architecture and how to build the expression trees.

– Christian S. Perone

Cite this article as: Christian S. Perone, "Gandiva, using LLVM and Arrow to JIT and evaluate Pandas expressions," in Terra Incognita, 19/01/2020, https://blog.christianperone.com/2020/01/gandiva-using-llvm-and-arrow-to-jit-and-evaluate-pandas-expressions/.

## PyTorch – Internal Architecture Tour

Update 28 Feb 2019: I added a new blog post with a slide deck containing the presentation I did for PyData Montreal.

## Introduction

This post is a tour around the PyTorch codebase, it is meant to be a guide for the architectural design of PyTorch and its internals. My main goal is to provide something useful for those who are interested in understanding what happens beyond the user-facing API and show something new beyond what was already covered in other tutorials.

Note: PyTorch build system uses code generation extensively so I won’t repeat here what was already described by others. If you’re interested in understanding how this works, please read the following tutorials:

## Short intro to Python extension objects in C/C++

As you probably know, you can extend Python using C and C++ and develop what is called as “extension”. All the PyTorch heavy work is implemented in C/C++ instead of pure-Python. To define a new Python object type in C/C++, you define a structure like this one example below (which is the base for the autograd Variable class):

// Python object that backs torch.autograd.Variable
struct THPVariable {
PyObject* backward_hooks;
};

As you can see, there is a macro at the beginning of the definition, called PyObject_HEAD, this macro’s goal is the standardization of Python objects and will expand to another structure that contains a pointer to a type object (which defines initialization methods, allocators, etc) and also a field with a reference counter.

There are two extra macros in the Python API called Py_INCREF() and Py_DECREF(), which are used to increment and decrement the reference counter of Python objects. Multiple entities can borrow or own a reference to other objects (the reference counter is increased), and only when this reference counter reaches zero (when all references get destroyed), Python will automatically delete the memory from that object using its garbage collector.

Funny fact: it is very common in many applications to use small integer numbers as indexing, counters, etc. For efficiency, the official CPython interpreter caches the integers from -5 up to 256. For that reason, the statement a = 200; b = 200; a is b will be True, while the statement a = 300; b = 300; a is b will be False.

## Zero-copy PyTorch Tensor to Numpy and vice-versa

PyTorch has its own Tensor representation, which decouples PyTorch internal representation from external representations. However, as it is very common, especially when data is loaded from a variety of sources, to have Numpy arrays everywhere, therefore we really need to make conversions between Numpy and PyTorch tensors. For that reason, PyTorch provides two methods called from_numpy() and numpy(), that converts a Numpy array to a PyTorch array and vice-versa, respectively. If we look the code that is being called to convert a Numpy array into a PyTorch tensor, we can get more insights on the PyTorch’s internal representation:

at::Tensor tensor_from_numpy(PyObject* obj) {
if (!PyArray_Check(obj)) {
throw TypeError("expected np.ndarray (got %s)", Py_TYPE(obj)->tp_name);
}

auto array = (PyArrayObject*)obj;
int ndim = PyArray_NDIM(array);
auto sizes = to_aten_shape(ndim, PyArray_DIMS(array));
auto strides = to_aten_shape(ndim, PyArray_STRIDES(array));
// NumPy strides use bytes. Torch strides use element counts.
auto element_size_in_bytes = PyArray_ITEMSIZE(array);
for (auto& stride : strides) {
stride /= element_size_in_bytes;
}

// (...) - omitted for brevity

void* data_ptr = PyArray_DATA(array);
auto& type = CPU(dtype_to_aten(PyArray_TYPE(array)));
Py_INCREF(obj);
return type.tensorFromBlob(data_ptr, sizes, strides, [obj](void* data) {
AutoGIL gil;
Py_DECREF(obj);
});
}

(code from tensor_numpy.cpp)

As you can see from this code, PyTorch is obtaining all information (array metadata) from Numpy representation and then creating its own. However, as you can note from the marked line 18, PyTorch is getting a pointer to the internal Numpy array raw data instead of copying it. This means that PyTorch will create a reference for this data, sharing the same memory region with the Numpy array object for the raw Tensor data.

There is also an important point here: when Numpy array object goes out of scope and get a zero reference count, it will be garbage collected and destroyed, that’s why there is an increment in the reference counting of the Numpy array object at line 20.

After this, PyTorch will create a new Tensor object from this Numpy data blob, and in the creation of this new Tensor it passes the borrowed memory data pointer, together with the memory size and strides as well as a function that will be used later by the Tensor Storage (we’ll discuss this in the next section) to release the data by decrementing the reference counting to the Numpy array object and let Python take care of this object life cycle.

The tensorFromBlob() method will create a new Tensor, but only after creating a new “Storage” for this Tensor. The storage is where the actual data pointer will be stored (and not in the Tensor structure itself). This takes us to the next section about Tensor Storages.

## Tensor Storage

The actual raw data of the Tensor is not directly kept in the Tensor structure, but on another structure called Storage, which in turn is part of the Tensor structure.

As we saw in the previous code from tensor_from_numpy(), there is a call for tensorFromBlob() that will create a Tensor from the raw data blob. This last function will call another function called storageFromBlob() that will, in turn, create a storage for this data according to its type. In the case of a CPU float type, it will return a new CPUFloatStorage instance.

The CPUFloatStorage is basically a wrapper with utility functions around the actual storage structure called THFloatStorage that we show below:

typedef struct THStorage
{
real *data;
ptrdiff_t size;
int refcount;
char flag;
THAllocator *allocator;
void *allocatorContext;
struct THStorage *view;
} THStorage;

(code from THStorage.h)

As you can see, the THStorage holds a pointer to the raw data, its size, flags and also an interesting field called allocator that we’ll soon discuss. It is also important to note that there is no metadata regarding on how to interpret the data inside the THStorage, this is due to the fact that the storage is “dumb” regarding of its contents and it is the Tensor responsibility to know how to “view” or interpret this data.

From this, you already probably realized that we can have multiple tensors pointing to the same storage but with different views of this data, and that’s why viewing a tensor with a different shape (but keeping the same number of elements) is so efficient. This Python code below shows that the data pointer in the storage is being shared after changing the way Tensor views its data:

>>> tensor_a = torch.ones((3, 3))
>>> tensor_b = tensor_a.view(9)
>>> tensor_a.storage().data_ptr() == tensor_b.storage().data_ptr()
True

As we can see in the example above, the data pointer on the storage of both Tensors are the same, but the Tensors represent a different interpretation of the storage data.

Now, as we saw in line 7 of the THFloatStorage structure, there is a pointer to a THAllocator structure there. And this is very important because it brings flexibility regarding the allocator that can be used to allocate the storage data. This structure is represented by the following code:

typedef struct THAllocator
{
void* (*malloc)(void*, ptrdiff_t);
void* (*realloc)(void*, void*, ptrdiff_t);
void (*free)(void*, void*);
} THAllocator;

(code from THAllocator.h)

As you can see, there are three function pointer fields in this structure to define what an allocator means: a malloc, realloc and free. For CPU-allocated memory, these functions will, of course, relate to the traditional malloc/realloc/free POSIX functions, however, when we want a storage allocated on GPUs we’ll end up using the CUDA allocators such as the cudaMallocHost(), like we can see in the THCudaHostAllocator malloc function below:

static void *THCudaHostAllocator_malloc(void* ctx, ptrdiff_t size) {
void* ptr;
if (size < 0) THError("Invalid memory size: %ld", size);
if (size == 0) return NULL;
THCudaCheck(cudaMallocHost(&ptr, size));
return ptr;
}

(code from THCAllocator.c)

You probably noticed a pattern in the repository organization, but it is important to keep in mind these conventions when navigating the repository, as summarized here (taken from the PyTorch lib readme):

• TH = TorcH
• THC = TorcH Cuda
• THCS = TorcH Cuda Sparse
• THCUNN = TorcH CUda Neural Network
• THD = TorcH Distributed
• THNN = TorcH Neural Network
• THS = TorcH Sparse

This convention is also present in the function/class names and other objects, so it is important to always keep these patterns in mind. While you can find CPU allocators in the TH code, you’ll find CUDA allocators in the THC code.

Finally, we can see the composition of the main Tensor THTensor structure:

typedef struct THTensor
{
int64_t *size;
int64_t *stride;
int nDimension;
THStorage *storage;
ptrdiff_t storageOffset;
int refcount;
char flag;
} THTensor;

(Code from THTensor.h)

And as you can see, the main THTensor structure holds the size/strides/dimensions/offsets/etc as well as the storage (THStorage) for the Tensor data.

We can summarize all this structure that we saw in the diagram below:

Now, once we have requirements such as multi-processing where we want to share tensor data among multiple different processes, we need a shared memory approach to solve it, otherwise, every time another process needs a tensor or even when you want to implement Hogwild training procedure where all different processes will write to the same memory region (where the parameters are), you’ll need to make copies between processes, and this is very inefficient. Therefore we’ll discuss in the next section a special kind of storage for Shared Memory.

## Shared Memory

Shared memory can be implemented in many different ways depending on the platform support. PyTorch supports some of them, but for the sake of simplicity, I’ll talk here about what happens on MacOS using the CPU (instead of GPU). Since PyTorch supports multiple shared memory approaches, this part is a little tricky to grasp into since it involves more levels of indirection in the code.

PyTorch provides a wrapper around the Python multiprocessing module and can be imported from torch.multiprocessing. The changes they implemented in this wrapper around the official Python multiprocessing were done to make sure that everytime a tensor is put on a queue or shared with another process, PyTorch will make sure that only a handle for the shared memory will be shared instead of a new entire copy of the Tensor.

Now, many people aren’t aware of a Tensor method from PyTorch called share_memory_(), however, this function is what triggers an entire rebuild of the storage memory for that particular Tensor. What this method does is to create a region of shared memory that can be used among different processes. This function will, in the end, call this following function below:

static THStorage* THPStorage_(newFilenameStorage)(ptrdiff_t size)
{
int flags = TH_ALLOCATOR_MAPPED_SHAREDMEM | TH_ALLOCATOR_MAPPED_EXCLUSIVE;
std::string handle = THPStorage_(__newHandle)();
auto ctx = libshm_context_new(NULL, handle.c_str(), flags);
return THStorage_(newWithAllocator)(size, &THManagedSharedAllocator, (void*)ctx);
}

(Code from StorageSharing.cpp)

And as you can see, this function will create another storage using a special allocator called THManagedSharedAllocator. This function first defines some flags and then it creates a handle which is a string in the format /torch_[process id]_[random number], and after that, it will then create a new storage using the special THManagedSharedAllocator. This allocator has function pointers to an internal PyTorch library called libshm, that will implement a Unix Domain Socket communication to share the shared memory region handles. This allocator is actual an especial case and it is a kind of “smart allocator” because it contains the communication control logic as well as it uses another allocator called THRefcountedMapAllocator that will be responsible for creating the actual shared memory region and call mmap() to map this region to the process virtual address space.

Note: when a method ends with a underscore in PyTorch, such as the method called share_memory_(), it means that this method has an in-place effect, and it will change the current object instead of creating a new one with the modifications.

I’ll now show a Python example of one processing using the data from a Tensor that was allocated on another process by manually exchanging the shared memory handle:

This is executed in the process A:

>>> import torch
>>> tensor_a = torch.ones((5, 5))
>>> tensor_a

1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
[torch.FloatTensor of size 5x5]

>>> tensor_a.is_shared()
False
>>> tensor_a = tensor_a.share_memory_()
>>> tensor_a.is_shared()
True
>>> tensor_a_storage = tensor_a.storage()
>>> tensor_a_storage._share_filename_()
(b'/var/tmp/tmp.0.yowqlr', b'/torch_31258_1218748506', 25)

In this code, executed in the process A, we create a new Tensor of 5×5 filled with ones. After that we make it shared and print the tuple with the Unix Domain Socket address as well as the handle. Now we can access this memory region from another process B as shown below:

Code executed in the process B:

>>> import torch
>>> tensor_a = torch.Tensor()
>>> tuple_info = (b'/var/tmp/tmp.0.yowqlr', b'/torch_31258_1218748506', 25)
>>> storage = torch.Storage._new_shared_filename(*tuple_info)
>>> tensor_a = torch.Tensor(storage).view((5, 5))

1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
1  1  1  1  1
[torch.FloatTensor of size 5x5]

As you can see, using the tuple information about the Unix Domain Socket address and the handle we were able to access the Tensor storage from another process. If you change the tensor in this process B, you’ll also see that it will reflect in the process A because these Tensors are sharing the same memory region.

## DLPack: a hope for the Deep Learning frameworks Babel

Now I would like to talk about something recent in the PyTorch code base, that is called DLPack. DLPack is an open standardization of an in-memory tensor structure that will allow exchange tensor data between frameworks, and what is quite interesting is that since this memory representation is standardized and very similar to the memory representation already in use by many frameworks, it will allow a zero-copy data sharing between frameworks, which is a quite amazing initiative given the variety of frameworks we have today without inter-communication among them.

This will certainly help to overcome the “island model” that we have today between tensor representations in MXNet, PyTorch, etc, and will allow developers to mix framework operations between frameworks and all the benefits that a standardization can bring to the frameworks.

The core of DLPack os a very simple structure called DLTensor, as shown below:

/*!
* \brief Plain C Tensor object, does not manage memory.
*/
typedef struct {
/*!
* \brief The opaque data pointer points to the allocated data.
*  This will be CUDA device pointer or cl_mem handle in OpenCL.
*  This pointer is always aligns to 256 bytes as in CUDA.
*/
void* data;
/*! \brief The device context of the tensor */
DLContext ctx;
/*! \brief Number of dimensions */
int ndim;
/*! \brief The data type of the pointer*/
DLDataType dtype;
/*! \brief The shape of the tensor */
int64_t* shape;
/*!
* \brief strides of the tensor,
*  can be NULL, indicating tensor is compact.
*/
int64_t* strides;
/*! \brief The offset in bytes to the beginning pointer to data */
uint64_t byte_offset;
} DLTensor;

(code from dlpack.h)

As you can see, there is a data pointer for the raw data as well as shape/stride/offset/GPU vs CPU, and other metadata information about the data that the DLTensor pointing to.

There is also a managed version of the tensor that is called DLManagedTensor, where the frameworks can provide a context and also a “deleter” function that can be called by the framework who borrowed the Tensor to inform the other framework that the resources are no longer required.

In PyTorch, if you want to convert to or from a DLTensor format, you can find both C/C++ methods for doing that or even in Python you can do that as shown below:

import torch
from torch.utils import dlpack

t = torch.ones((5, 5))
dl = dlpack.to_dlpack(t)

This Python function will call the toDLPack function from ATen, shown below:

DLManagedTensor* toDLPack(const Tensor& src) {
ATenDLMTensor * atDLMTensor(new ATenDLMTensor);
atDLMTensor->handle = src;
atDLMTensor->tensor.manager_ctx = atDLMTensor;
atDLMTensor->tensor.deleter = &deleter;
atDLMTensor->tensor.dl_tensor.data = src.data_ptr();
int64_t device_id = 0;
if (src.type().is_cuda()) {
device_id = src.get_device();
}
atDLMTensor->tensor.dl_tensor.ctx = getDLContext(src.type(), device_id);
atDLMTensor->tensor.dl_tensor.ndim = src.dim();
atDLMTensor->tensor.dl_tensor.dtype = getDLDataType(src.type());
atDLMTensor->tensor.dl_tensor.shape = const_cast<int64_t*>(src.sizes().data());
atDLMTensor->tensor.dl_tensor.strides = const_cast<int64_t*>(src.strides().data());
atDLMTensor->tensor.dl_tensor.byte_offset = 0;
return &(atDLMTensor->tensor);
}

As you can see, it’s a pretty simple conversion, casting the metadata from the PyTorch format to the DLPack format and assigning a pointer to the internal Tensor data representation.

I really hope that more frameworks adopt this standard that will certainly give benefits to the ecosystem. It is also interesting to note that a potential integration with Apache Arrow would be amazing.

That’s it, I hope you liked this long post !

– Christian S. Perone

Cite this article as: Christian S. Perone, "PyTorch – Internal Architecture Tour," in Terra Incognita, 12/03/2018, https://blog.christianperone.com/2018/03/pytorch-internal-architecture-tour/.

## Privacy-preserving sentence semantic similarity using InferSent embeddings and secure two-party computation

### Privacy-preserving Computation

Privacy-preserving computation or secure computation is a sub-field of cryptography where two (two-party, or 2PC) or multiple (multi-party, or MPC) parties can evaluate a function together without revealing information about the parties private input data to each other. The problem and the first solution to it were introduced in 1982 by an amazing breakthrough done by Andrew Yao on what later became known as the “Yao’s Millionaires’ problem“.

The Yao’s Millionaires Problem is where two millionaires, Alice and Bob, who are interested in knowing which of them is richer but without revealing to each other their actual wealth. In other words, what they want can be generalized as that: Alice and Bob want jointly compute a function securely, without knowing anything other than the result of the computation on the input data (that remains private to them).

To make the problem concrete, Alice has an amount A such as $10, and Bob has an amount B such as$ 50, and what they want to know is which one is larger, without Bob revealing the amount B to Alice or Alice revealing the amount A to Bob. It is also important to note that we also don’t want to trust on a third-party, otherwise the problem would just be a simple protocol of information exchange with the trusted party.

Formally what we want is to jointly evaluate the following function:

$r = f(A, B)$

Such as the private values A and B are held private to the sole owner of it and where the result r will be known to just one or both of the parties.

It seems very counterintuitive that a problem like that could ever be solved, but for the surprise of many people, it is possible to solve it on some security requirements. Thanks to the recent developments in techniques such as FHE (Fully Homomorphic Encryption), Oblivious Transfer, Garbled Circuits, problems like that started to get practical for real-life usage and they are being nowadays being used by many companies in applications such as information exchange, secure location, advertisement, satellite orbit collision avoidance, etc.

I’m not going to enter into details of these techniques, but if you’re interested in the intuition behind the OT (Oblivious Transfer), you should definitely read the amazing explanation done by Craig Gidney here. There are also, of course, many different protocols for doing 2PC or MPC, where each one of them assumes some security requirements (semi-honest, malicious, etc), I’m not going to enter into the details to keep the post focused on the goal, but you should be aware of that.

### The problem: sentence similarity

What we want to achieve is to use privacy-preserving computation to calculate the similarity between sentences without disclosing the content of the sentences. Just to give a concrete example: Bob owns a company and has the description of many different projects in sentences such as: “This project is about building a deep learning sentiment analysis framework that will be used for tweets“, and Alice who owns another competitor company, has also different projects described in similar sentences. What they want to do is to jointly compute the similarity between projects in order to find if they should be doing partnership on a project or not, however, and this is the important point: Bob doesn’t want Alice to know the project descriptions and neither Alice wants Bob to be aware of their projects, they want to know the closest match between the different projects they run, but without disclosing the project ideas (project descriptions).

### Sentence Similarity Comparison

Now, how can we exchange information about the Bob and Alice’s project sentences without disclosing information about the project descriptions ?

One naive way to do that would be to just compute the hashes of the sentences and then compare only the hashes to check if they match. However, this would assume that the descriptions are exactly the same, and besides that, if the entropy of the sentences is small (like small sentences), someone with reasonable computation power can try to recover the sentence.

Another approach for this problem (this is the approach that we’ll be using), is to compare the sentences in the sentence embeddings space. We just need to create sentence embeddings using a Machine Learning model (we’ll use InferSent later) and then compare the embeddings of the sentences. However, this approach also raises another concern: what if Bob or Alice trains a Seq2Seq model that would go from the embeddings of the other party back to an approximate description of the project ?

It isn’t unreasonable to think that one can recover an approximate description of the sentence given their embeddings. That’s why we’ll use the two-party secure computation for computing the embeddings similarity, in a way that Bob and Alice will compute the similarity of the embeddings without revealing their embeddings, keeping their project ideas safe.

The entire flow is described in the image below, where Bob and Alice shares the same Machine Learning model, after that they use this model to go from sentences to embeddings, followed by a secure computation of the similarity in the embedding space.

### Generating sentence embeddings with InferSent

InferSent is an NLP technique for universal sentence representation developed by Facebook that uses supervised training to produce high transferable representations.

They used a Bi-directional LSTM with attention that consistently surpassed many unsupervised training methods such as the SkipThought vectors. They also provide a Pytorch implementation that we’ll use to generate sentence embeddings.

Note: even if you don’t have GPU, you can have reasonable performance doing embeddings for a few sentences.

The first step to generate the sentence embeddings is to download and load a pre-trained InferSent model:

import numpy as np
import torch

GLOVE_EMBS = '../dataset/GloVe/glove.840B.300d.txt'
INFERSENT_MODEL = 'infersent.allnli.pickle'

map_location=lambda storage, loc: storage)

model.set_glove_path(GLOVE_EMBS)
model.build_vocab_k_words(K=100000)

Now we need to define a similarity measure to compare two vectors, and for that goal, I’ll the cosine similarity (I wrote a tutorial about this similarity measure here) since it’s pretty straightforward:

$cos(\pmb x, \pmb y) = \frac {\pmb x \cdot \pmb y}{||\pmb x|| \cdot ||\pmb y||}$

As you can see, if we have two unit vectors (vectors with norm 1), the two terms in the equation denominator will be 1 and we will be able to remove the entire denominator of the equation, leaving only:

$cos(\hat{x}, \hat{y}) =\hat{x} \cdot\hat{y}$

So, if we normalize our vectors to have a unit norm (that’s why the vectors are wearing hats in the equation above), we can make the computation of the cosine similarity become just a simple dot product. That will help us a lot in computing the similarity distance later when we’ll use a framework to do the secure computation of this dot product.

So, the next step is to define a function that will take some sentence text and forward it to the model to generate the embeddings and then normalize them to unit vectors:

# This function will forward the text into the model and
# get the embeddings. After that, it will normalize it
# to a unit vector.

def encode(model, text):
embedding = model.encode([text])[0]
embedding /= np.linalg.norm(embedding)
return embedding

As you can see, this function is pretty simple, it feeds the text into the model, and then it will divide the embedding vector by the embedding norm.

Now, for practical reasons, I’ll be using integer computation later for computing the similarity, however, the embeddings generated by InferSent are of course real values. For that reason, you’ll see in the code below that we create another function to scale the float values and remove the radix point and converting them to integers. There is also another important issue, the framework that we’ll be using later for secure computation doesn’t allow signed integers, so we also need to clip the embeddings values between 0.0 and 1.0. This will of course cause some approximation errors, however, we can still get very good approximations after clipping and scaling with limited precision (I’m using 14 bits for scaling to avoid overflow issues later during dot product computations):

# This function will scale the embedding in order to

def scale(embedding):
SCALE = 1 << 14
scale_embedding = np.clip(embedding, 0.0, 1.0) * SCALE
return scale_embedding.astype(np.int32)


You can use floating-point in your secure computations and there are a lot of frameworks that support them, however, it is more tricky to do that, and for that reason, I used integer arithmetic to simplify the tutorial. The function above is just a hack to make it simple. It’s easy to see that we can recover this embedding later without too much loss of precision.

Now we just need to create some sentence samples that we’ll be using:

# The list of Alice sentences
alice_sentences = [
'my cat loves to walk over my keyboard',
'I like to pet my cat',
]

# The list of Bob sentences
bob_sentences = [
'the cat is always walking over my keyboard',
]

And convert them to embeddings:

# Alice sentences
alice_sentence1 = encode(model, alice_sentences[0])
alice_sentence2 = encode(model, alice_sentences[1])

# Bob sentences
bob_sentence1 = encode(model, bob_sentences[0])

Since we have now the sentences and every sentence is also normalized, we can compute cosine similarity just by doing a dot product between the vectors:

>>> np.dot(bob_sentence1, alice_sentence1)
0.8798542

>>> np.dot(bob_sentence1, alice_sentence2)
0.62976325

As we can see, the first sentence of Bob is most similar (~0.87) with Alice first sentence than to the Alice second sentence (~0.62).

Since we have now the embeddings, we just need to convert them to scaled integers:

# Scale the Alice sentence embeddings
alice_sentence1_scaled = scale(alice_sentence1)
alice_sentence2_scaled = scale(alice_sentence2)

# Scale the Bob sentence embeddings
bob_sentence1_scaled = scale(bob_sentence1)

# This is the unit vector embedding for the sentence
>>> alice_sentence1
array([ 0.01698913, -0.0014404 ,  0.0010993 , ...,  0.00252409,
0.00828147,  0.00466533], dtype=float32)

# This is the scaled vector as integers
>>> alice_sentence1_scaled
array([278,   0,  18, ...,  41, 135,  76], dtype=int32)

Now with these embeddings as scaled integers, we can proceed to the second part, where we’ll be doing the secure computation between two parties.

### Two-party secure computation

In order to perform secure computation between the two parties (Alice and Bob), we’ll use the ABY framework. ABY implements many difference secure computation schemes and allows you to describe your computation as a circuit like pictured in the image below, where the Yao’s Millionaire’s problem is described:

As you can see, we have two inputs entering in one GT GATE (greater than gate) and then a output. This circuit has a bit length of 3 for each input and will compute if the Alice input is greater than (GT GATE) the Bob input. The computing parties then secret share their private data and then can use arithmetic sharing, boolean sharing, or Yao sharing to securely evaluate these gates.

ABY is really easy to use because you can just describe your inputs, shares, gates and it will do the rest for you such as creating the socket communication channel, exchanging data when needed, etc. However, the implementation is entirely written in C++ and I’m not aware of any Python bindings for it (a great contribution opportunity).

Fortunately, there is an implemented example for ABY that can do dot product calculation for us, the example is here. I won’t replicate the example here, but the only part that we have to change is to read the embedding vectors that we created before instead of generating random vectors and increasing the bit length to 32-bits.

After that, we just need to execute the application on two different machines (or by emulating locally like below):

# This will execute the server part, the -r 0 specifies the role (server)
# and the -n 4096 defines the dimension of the vector (InferSent generates
# 4096-dimensional embeddings).
~# ./innerproduct -r 0 -n 4096

# And the same on another process (or another machine, however for another
# machine execution you'll have to obviously specify the IP).
~# ./innerproduct -r 1 -n 4096

And we get the following results:

Inner Product of alice_sentence1 and bob_sentence1  = 226691917
Inner Product of alice_sentence2 and bob_sentence1  = 171746521

Even in the integer representation, you can see that the inner product of the Alice’s first sentence and the Bob sentence is higher, meaning that the similarity is also higher. But let’s now convert this value back to float:

>>> SCALE = 1 << 14

# This is the dot product we should get
>>> np.dot(alice_sentence1, bob_sentence1)
0.8798542

# This is the inner product we got on secure computation
>>> 226691917 / SCALE**2.0
0.8444931

# This is the dot product we should get
>>> np.dot(alice_sentence2, bob_sentence1)
0.6297632

# This is the inner product we got on secure computation
>>> 171746521 / SCALE**2.0
0.6398056

As you can see, we got very good approximations, even in presence of low-precision math and unsigned integer requirements. Of course that in real-life you won’t have the two values and vectors, because they’re supposed to be hidden, but the changes to accommodate that are trivial, you just need to adjust ABY code to load only the vector of the party that it is executing it and using the correct IP addresses/port of the both parties.

I hope you liked it !

– Christian S. Perone

Cite this article as: Christian S. Perone, "Privacy-preserving sentence semantic similarity using InferSent embeddings and secure two-party computation," in Terra Incognita, 22/01/2018, https://blog.christianperone.com/2018/01/privacy-preserving-infersent/.

## Nanopipe: connecting the modern babel

For more information, see official documentation site or the official Github repository.

Hello everyone, I just released the Nanopipe project. Nanopipe is a library that allows you to connect different message queue systems (but not limited to) together. Nanopipe was built to avoid the glue code between different types of communication protocols/channels that is very common nowadays. An example of this is: you have an application that is listening for messages on an AMQP broker (ie. RabbitMQ) but you also have a Redis pub/sub source of messages and also a MQTT source from a weird IoT device you may have. Using Nanopipe, you can connect both MQTT and Redis to RabbitMQ without doing any glue code for that. You can also build any kind of complex connection scheme using Nanopipe.

## JIT native code generation for TensorFlow computation graphs using Python and LLVM

Update: Hacker News discussion here.

### The TensorFlow Computation Graph

One of the most amazing components of the TensorFlow architecture is the computation graph that can be serialized using Protocol Buffers. This computation graph follows a well-defined format (click here for the proto files) and describes the computation that you specify (it can be a Deep Learning model like a CNN, a simple Logistic Regression or even any computation you want). For instance, here is an example of a very simple TensorFlow computation graph that we will use in this tutorial (using TensorFlow Python API):

import tensorflow as tf

with tf.Session() as sess:
input_placeholder = tf.placeholder(tf.int32, 1, name="input")
sub_op = tf.sub(input_placeholder, tf.constant(2, dtype=tf.int32))
name="output")
tf.train.write_graph(sess.graph_def, ".", "graph.pb", True)


As you can see, this is a very simple computation graph. First, we define the placeholder that will hold the input tensor and after that we specify the computation that should happen using this input tensor as input data. Here we can also see that we’re defining two important nodes of this graph, one is called “input” (the aforementioned placeholder) and the other is called “output“, that will hold the result of the final computation. This graph is the same as the following formula for a scalar: $output = (((input - 2)-5)+100)$, where I intentionally added redundant operations to see LLVM constant propagation later.

In the last line of the code, we’re persisting this computation graph (including the constant values) into a serialized protobuf file. The final True parameter is to output a textual representation instead of binary, so it will produce the following human-readable output protobuf file (I omitted a part of it for brevity):

node {
name: "input"
op: "Placeholder"
attr {
key: "dtype"
value {
type: DT_INT32
}
}
attr {
key: "shape"
value {
shape {
dim {
size: 1
}
}
}
}
}
node {
name: "Const"
op: "Const"
attr {
key: "dtype"
value {
type: DT_INT32
}
}
attr {
key: "value"
value {
tensor {
dtype: DT_INT32
tensor_shape {
}
int_val: 2
}
}
}
}

--- >(omitted for brevity) < ---

node {
name: "output"
input: "Const_2"
attr {
key: "T"
value {
type: DT_INT32
}
}
}
versions {
producer: 9
}


This is a very simple graph, and TensorFlow graphs are actually never that simple, because TensorFlow models can easily contain more than 300 nodes depending on the model you’re specifying, specially for Deep Learning models.

We’ll use the above graph to show how we can JIT native code for this simple graph using LLVM framework.

### The LLVM Frontend, IR and Backend

The LLVM framework is a really nice, modular and complete ecosystem for building compilers and toolchains. A very nice description of the LLVM architecture that is important for us is shown in the picture below:

(The picture above is just a small part of the LLVM architecture, for a comprehensive description of it, please see the nice article from the AOSA book written by Chris Lattner)

Looking in the image above, we can see that LLVM provides a lot of core functionality, in the left side you see that many languages can write code for their respective language frontends, after that it doesn’t matter in which language you wrote your code, everything is transformed into a very powerful language called LLVM IR (LLVM Intermediate Representation) which is as you can imagine, a intermediate representation of the code just before the assembly code itself. In my opinion, the IR is the key component of what makes LLVM so amazing, because it doesn’t matter in which language you wrote your code (or even if it was a JIT’ed IR), everything ends in the same representation, and then here is where the magic happens, because the IR can take advantage of the LLVM optimizations (also known as transform and analysis passes).

After this IR generation, you can feed it into any LLVM backend to generate native code for any architecture supported by LLVM (such as x86, ARM, PPC, etc) and then you can finally execute your code with the native performance and also after LLVM optimization passes.

In order to JIT code using LLVM, all you need is to build the IR programmatically, create a execution engine to convert (during execution-time) the IR into native code, get a pointer for the function you have JIT’ed and then finally execute it. I’ll use here a Python binding for LLVM called llvmlite, which is very Pythonic and easy to use.

### JIT’ing TensorFlow Graph using Python and LLVM

Let’s now use the LLVM and Python to JIT the TensorFlow computational graph. This is by no means a comprehensive implementation, it is very simplistic approach, a oversimplification that assumes some things: a integer closure type, just some TensorFlow operations and also a single scalar support instead of high rank tensors.

So, let’s start building our JIT code; first of all, let’s import the required packages, initialize some LLVM sub-systems and also define the LLVM respective type for the TensorFlow integer type:

from ctypes import CFUNCTYPE, c_int

import tensorflow as tf
from tensorflow.core.framework import graph_pb2
from tensorflow.core.framework import types_pb2
from tensorflow.python.framework import ops

import llvmlite.ir as ll
import llvmlite.binding as llvm

llvm.initialize()
llvm.initialize_native_target()
llvm.initialize_native_asmprinter()

TYPE_TF_LLVM = {
types_pb2.DT_INT32: ll.IntType(32),
}

After that, let’s define a class to open the TensorFlow exported graph and also declare a method to get a node of the graph by name:

class TFGraph(object):
def __init__(self, filename="graph.pb", binary=False):
self.graph_def = graph_pb2.GraphDef()
with open("graph.pb", "rb") as f:
if binary:
else:

def get_node(self, name):
for node in self.graph_def.node:
if node.name == name:
return node

And let’s start by defining our main function that will be the starting point of the code:

def run_main():
graph = TFGraph("graph.pb", False)
input_node = graph.get_node("input")
output_node = graph.get_node("output")

input_type = TYPE_TF_LLVM[input_node.attr["dtype"].type]
output_type = TYPE_TF_LLVM[output_node.attr["T"].type]

module = ll.Module()
func_type = ll.FunctionType(output_type, [input_type])
func = ll.Function(module, func_type, name='tensorflow_graph')
func.args[0].name = 'input'

bb_entry = func.append_basic_block('entry')
ir_builder = ll.IRBuilder(bb_entry)

As you can see in the code above, we open the serialized protobuf graph and then get the input and output nodes of this graph. After that we also map the type of the both graph nodes (input/output) to the LLVM type (from TensorFlow integer to LLVM integer). We start then by defining a LLVM Module, which is the top level container for all IR objects. One module in LLVM can contain many different functions, here we will create just one function that will represent the graph, this function will receive as input argument the input data of the same type of the input node and then it will return a value with the same type of the output node.

After that we start by creating the entry block of the function and using this block we instantiate our IR Builder, which is a object that will provide us the building blocks for JIT’ing operations of TensorFlow graph.

Let’s now define the function that will do the real work of converting TensorFlow nodes into LLVM IR:

def build_graph(ir_builder, graph, node):
left_op_node = graph.get_node(node.input[0])
right_op_node = graph.get_node(node.input[1])
left_op = build_graph(ir_builder, graph, left_op_node)
right_op = build_graph(ir_builder, graph, right_op_node)

if node.op == "Sub":
left_op_node = graph.get_node(node.input[0])
right_op_node = graph.get_node(node.input[1])
left_op = build_graph(ir_builder, graph, left_op_node)
right_op = build_graph(ir_builder, graph, right_op_node)
return ir_builder.sub(left_op, right_op)

if node.op == "Placeholder":
function_args = ir_builder.function.args
for arg in function_args:
if arg.name == node.name:
return arg

if node.op == "Const":
llvm_const_type = TYPE_TF_LLVM[node.attr["dtype"].type]
const_value = node.attr["value"].tensor.int_val[0]
llvm_const_value = llvm_const_type(const_value)
return llvm_const_value

In this function, we receive by parameters the IR Builder, the graph class that we created earlier and the output node. This function will then recursively build the LLVM IR by means of the IR Builder. Here you can see that I only implemented the Add/Sub/Placeholder and Const operations from the TensorFlow graph, just to be able to support the graph that we defined earlier.

After that, we just need to define a function that will take a LLVM Module and then create a execution engine that will execute the LLVM optimization over the LLVM IR before doing the hard-work of converting the IR into native x86 code:

def create_engine(module):
features = llvm.get_host_cpu_features().flatten()
llvm_module = llvm.parse_assembly(str(module))
target = llvm.Target.from_default_triple()
target_machine = target.create_target_machine(opt=3, features=features)
engine = llvm.create_mcjit_compiler(llvm_module, target_machine)
engine.finalize_object()
print target_machine.emit_assembly(llvm_module)
return engine

In the code above, you can see that we first get the CPU features (SSE, etc) into a list, after that we parse the LLVM IR from the module and then we create a engine using maximum optimization level (opt=3, roughly equivalent to the GCC -O3 parameter), we’re also printing the assembly code (in my case, the x86 assembly built by LLVM).

And here we just finish our run_main() function:

ret = build_graph(ir_builder, graph, output_node)
ir_builder.ret(ret)

with open("output.ir", "w") as f:
f.write(str(module))

engine = create_engine(module)

cfunc = CFUNCTYPE(c_int, c_int)(func_ptr)
ret = cfunc(10)

print "Execution output: {}".format(ret)

As you can see in the code above, we just call the build_graph() method and then use the IR Builder to add the “ret” LLVM IR instruction (ret = return) to return the output of the IR function we just created based on the TensorFlow graph. We’re also here writing the IR output to a external file, I’ll use this LLVM IR file later to create native assembly for other different architectures such as ARM architecture. And finally, just get the native code function address, create a Python wrapper for this function and then call it with the argument “10”, which will be input data and then output the resulting output value.

And that is it, of course that this is just a oversimplification, but now we understand the advantages of having a JIT for our TensorFlow models.

### The output LLVM IR, the advantage of optimizations and multiple architectures (ARM, PPC, x86, etc)

For instance, lets create the LLVM IR (using the code I shown above) of the following TensorFlow graph:

import tensorflow as tf

with tf.Session() as sess:
input_placeholder = tf.placeholder(tf.int32, 1, name="input")
sub_op = tf.sub(input_placeholder, tf.constant(2, dtype=tf.int32))
name="output")
tf.train.write_graph(sess.graph_def, ".", "graph.pb", True)

The LLVM IR generated is this one below:

; ModuleID = ""
target triple = "unknown-unknown-unknown"
target datalayout = ""

define i32 @"tensorflow_graph"(i32 %"input")
{
entry:
%".3" = sub i32 %"input", 2
%".4" = add i32 %".3", 5
%".5" = add i32 %".4", 100
ret i32 %".5"
}


As you can see, the LLVM IR looks a lot like an assembly code, but this is not the final assembly code, this is just a non-optimized IR yet. Just before generating the x86 assembly code, LLVM runs a lot of optimization passes over the LLVM IR, and it will do things such as dead code elimination, constant propagation, etc. And here is the final native x86 assembly code that LLVM generates for the above LLVM IR of the TensorFlow graph:

    .text
.file	"<string>"
.globl	tensorflow_graph
.align	16, 0x90
.type	tensorflow_graph,@function
tensorflow_graph:
.cfi_startproc
leal	103(%rdi), %eax
retq
.Lfunc_end0:
.size	tensorflow_graph, .Lfunc_end0-tensorflow_graph
.cfi_endproc

.section	".note.GNU-stack","",@progbits

As you can see, the optimized code removed a lot of redundant operations, and ended up just doing a add operation of 103, which is the correct simplification of the computation that we defined in the graph. For large graphs, you can see that these optimizations can be really powerful, because we are reusing the compiler optimizations that were developed for years in our Machine Learning model computation.

You can also use a LLVM tool called “llc”, that can take an LLVM IR file and the generate assembly for any other platform you want, for instance, the command-line below will generate native code for ARM architecture:

llc -O3 out.ll -march=arm -o sample.s

The output sample.s file is the one below:

    .text
.syntax unified
.eabi_attribute	67, "2.09"	@ Tag_conformance
.eabi_attribute	6, 1	@ Tag_CPU_arch
.eabi_attribute	8, 1	@ Tag_ARM_ISA_use
.eabi_attribute	17, 1	@ Tag_ABI_PCS_GOT_use
.eabi_attribute	20, 1	@ Tag_ABI_FP_denormal
.eabi_attribute	21, 1	@ Tag_ABI_FP_exceptions
.eabi_attribute	23, 3	@ Tag_ABI_FP_number_model
.eabi_attribute	34, 1	@ Tag_CPU_unaligned_access
.eabi_attribute	24, 1	@ Tag_ABI_align_needed
.eabi_attribute	25, 1	@ Tag_ABI_align_preserved
.eabi_attribute	38, 1	@ Tag_ABI_FP_16bit_format
.eabi_attribute	14, 0	@ Tag_ABI_PCS_R9_use
.file	"out.ll"
.globl	tensorflow_graph
.align	2
.type	tensorflow_graph,%function
tensorflow_graph:                       @ @tensorflow_graph
.fnstart
@ BB#0:                                 @ %entry
mov	pc, lr
.Lfunc_end0:
.size	tensorflow_graph, .Lfunc_end0-tensorflow_graph
.fnend

.section	".note.GNU-stack","",%progbits

As you can see above, the ARM assembly code is also just a “add” assembly instruction followed by a return instruction.

This is really nice because we can take natural advantage of the LLVM framework. For instance, today ARM just announced the ARMv8-A with Scalable Vector Extensions (SVE) that will support 2048-bit vectors, and they are already working on patches for LLVM. In future, a really nice addition to LLVM would be the development of LLVM Passes for analysis and transformation that would take into consideration the nature of Machine Learning models.

And that’s it, I hope you liked the post ! Is really awesome what you can do with a few lines of Python, LLVM and TensorFlow.

Update 22 Aug 2016: Josh Klontz just pointed his cool project called Likely on Hacker News discussion.

Update 22 Aug 2016: TensorFlow team is actually working on a JIT (I don’t know if they are using LLVM, but it seems the most reasonable way to go in my opinion). In their paper, there is also a very important statement regarding Future Work that I cite here:

“We also have a number of concrete directions to improve the performance of TensorFlow. One such direction is our initial work on a just-in-time compiler that can take a subgraph of a TensorFlow execution, perhaps with some runtime profiling information about the typical sizes and shapes of tensors, and can generate an optimized routine for this subgraph. This compiler will understand the semantics of perform a number of optimizations such as loop fusion, blocking and tiling for locality, specialization for particular shapes and sizes, etc.” – TensorFlow White Paper

### Full code

from ctypes import CFUNCTYPE, c_int

import tensorflow as tf
from tensorflow.core.framework import graph_pb2
from tensorflow.core.framework import types_pb2
from tensorflow.python.framework import ops

import llvmlite.ir as ll
import llvmlite.binding as llvm

llvm.initialize()
llvm.initialize_native_target()
llvm.initialize_native_asmprinter()

TYPE_TF_LLVM = {
types_pb2.DT_INT32: ll.IntType(32),
}

class TFGraph(object):
def __init__(self, filename="graph.pb", binary=False):
self.graph_def = graph_pb2.GraphDef()
with open("graph.pb", "rb") as f:
if binary:
else:

def get_node(self, name):
for node in self.graph_def.node:
if node.name == name:
return node

def build_graph(ir_builder, graph, node):
left_op_node = graph.get_node(node.input[0])
right_op_node = graph.get_node(node.input[1])
left_op = build_graph(ir_builder, graph, left_op_node)
right_op = build_graph(ir_builder, graph, right_op_node)

if node.op == "Sub":
left_op_node = graph.get_node(node.input[0])
right_op_node = graph.get_node(node.input[1])
left_op = build_graph(ir_builder, graph, left_op_node)
right_op = build_graph(ir_builder, graph, right_op_node)
return ir_builder.sub(left_op, right_op)

if node.op == "Placeholder":
function_args = ir_builder.function.args
for arg in function_args:
if arg.name == node.name:
return arg

if node.op == "Const":
llvm_const_type = TYPE_TF_LLVM[node.attr["dtype"].type]
const_value = node.attr["value"].tensor.int_val[0]
llvm_const_value = llvm_const_type(const_value)
return llvm_const_value

def create_engine(module):
features = llvm.get_host_cpu_features().flatten()
llvm_module = llvm.parse_assembly(str(module))
target = llvm.Target.from_default_triple()
target_machine = target.create_target_machine(opt=3, features=features)
engine = llvm.create_mcjit_compiler(llvm_module, target_machine)
engine.finalize_object()
print target_machine.emit_assembly(llvm_module)
return engine

def run_main():
graph = TFGraph("graph.pb", False)
input_node = graph.get_node("input")
output_node = graph.get_node("output")

input_type = TYPE_TF_LLVM[input_node.attr["dtype"].type]
output_type = TYPE_TF_LLVM[output_node.attr["T"].type]

module = ll.Module()
func_type = ll.FunctionType(output_type, [input_type])
func = ll.Function(module, func_type, name='tensorflow_graph')
func.args[0].name = 'input'

bb_entry = func.append_basic_block('entry')
ir_builder = ll.IRBuilder(bb_entry)

ret = build_graph(ir_builder, graph, output_node)
ir_builder.ret(ret)

with open("output.ir", "w") as f:
f.write(str(module))

engine = create_engine(module)

cfunc = CFUNCTYPE(c_int, c_int)(func_ptr)
ret = cfunc(10)

print "Execution output: {}".format(ret)

if __name__ == "__main__":
run_main()
Cite this article as: Christian S. Perone, "JIT native code generation for TensorFlow computation graphs using Python and LLVM," in Terra Incognita, 22/08/2016, https://blog.christianperone.com/2016/08/jit-native-code-generation-for-tensorflow-computation-graphs-using-python-and-llvm/.

## Deep learning – Convolutional neural networks and feature extraction with Python

Convolutional neural networks (or ConvNets) are biologically-inspired variants of MLPs, they have different kinds of layers and each different layer works different than the usual MLP layers. If you are interested in learning more about ConvNets, a good course is the CS231n – Convolutional Neural Newtorks for Visual Recognition. The architecture of the CNNs are shown in the images below:

As you can see, the ConvNets works with 3D volumes and transformations of these 3D volumes. I won’t repeat in this post the entire CS231n tutorial, so if you’re really interested, please take time to read before continuing.

### Lasagne and nolearn

One of the Python packages for deep learning that I really like to work with is Lasagne and nolearn. Lasagne is based on Theano so the GPU speedups will really make a great difference, and their declarative approach for the neural networks creation are really helpful. The nolearn libary is a collection of utilities around neural networks packages (including Lasagne) that can help us a lot during the creation of the neural network architecture, inspection of the layers, etc.

What I’m going to show in this post, is how to build a simple ConvNet architecture with some convolutional and pooling layers. I’m also going to show how you can use a ConvNet to train a feature extractor and then use it to extract features before feeding them into different models like SVM, Logistic Regression, etc. Many people use pre-trained ConvNet models and then remove the last output layer to extract the features from ConvNets that were trained on ImageNet datasets. This is usually called transfer learning because you can use layers from other ConvNets as feature extractors for different problems, since the first layer filters of the ConvNets works as edge detectors, they can be used as general feature detectors for other problems.

The MNIST dataset is one of the most traditional datasets for digits classification. We will use a pickled version of it for Python, but first, lets import the packages that we will need to use:

import matplotlib
import matplotlib.pyplot as plt
import matplotlib.cm as cm

from urllib import urlretrieve
import cPickle as pickle
import os
import gzip

import numpy as np
import theano

import lasagne
from lasagne import layers

from nolearn.lasagne import NeuralNet
from nolearn.lasagne import visualize

from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix

As you can see, we are importing matplotlib for plotting some images, some native Python modules to download the MNIST dataset, numpy, theano, lasagne, nolearn and some scikit-learn functions for model evaluation.

After that, we define our MNIST loading function (this is pretty the same function used in the Lasagne tutorial):

def load_dataset():
url = 'http://deeplearning.net/data/mnist/mnist.pkl.gz'
filename = 'mnist.pkl.gz'
if not os.path.exists(filename):
urlretrieve(url, filename)

with gzip.open(filename, 'rb') as f:

X_train, y_train = data[0]
X_val, y_val = data[1]
X_test, y_test = data[2]

X_train = X_train.reshape((-1, 1, 28, 28))
X_val = X_val.reshape((-1, 1, 28, 28))
X_test = X_test.reshape((-1, 1, 28, 28))

y_train = y_train.astype(np.uint8)
y_val = y_val.astype(np.uint8)
y_test = y_test.astype(np.uint8)

return X_train, y_train, X_val, y_val, X_test, y_test

As you can see, we are downloading the MNIST pickled dataset and then unpacking it into the three different datasets: train, validation and test. After that we reshape the image contents to prepare them to input into the Lasagne input layer later and we also convert the numpy array types to uint8 due to the GPU/theano datatype restrictions.

After that, we’re ready to load the MNIST dataset and inspect it:

X_train, y_train, X_val, y_val, X_test, y_test = load_dataset()
plt.imshow(X_train[0][0], cmap=cm.binary)

This code above will output the following image (I’m using IPython Notebook):

### ConvNet Architecture and Training

Now we can define our ConvNet architecture and then train it using a GPU/CPU (I have a very cheap GPU, but it helps a lot):

net1 = NeuralNet(
layers=[('input', layers.InputLayer),
('conv2d1', layers.Conv2DLayer),
('maxpool1', layers.MaxPool2DLayer),
('conv2d2', layers.Conv2DLayer),
('maxpool2', layers.MaxPool2DLayer),
('dropout1', layers.DropoutLayer),
('dense', layers.DenseLayer),
('dropout2', layers.DropoutLayer),
('output', layers.DenseLayer),
],
# input layer
input_shape=(None, 1, 28, 28),
# layer conv2d1
conv2d1_num_filters=32,
conv2d1_filter_size=(5, 5),
conv2d1_nonlinearity=lasagne.nonlinearities.rectify,
conv2d1_W=lasagne.init.GlorotUniform(),
# layer maxpool1
maxpool1_pool_size=(2, 2),
# layer conv2d2
conv2d2_num_filters=32,
conv2d2_filter_size=(5, 5),
conv2d2_nonlinearity=lasagne.nonlinearities.rectify,
# layer maxpool2
maxpool2_pool_size=(2, 2),
# dropout1
dropout1_p=0.5,
# dense
dense_num_units=256,
dense_nonlinearity=lasagne.nonlinearities.rectify,
# dropout2
dropout2_p=0.5,
# output
output_nonlinearity=lasagne.nonlinearities.softmax,
output_num_units=10,
# optimization method params
update=nesterov_momentum,
update_learning_rate=0.01,
update_momentum=0.9,
max_epochs=10,
verbose=1,
)

# Train the network
nn = net1.fit(X_train, y_train)

As you can see, in the parameter layers we’re defining a dictionary of tuples with the layer names/types and then we define the parameters for these layers. Our architecture here is using two convolutional layers with poolings and then a fully connected layer (dense layer) and the output layer. There are also dropouts between some layers, the dropout layer is a regularizer that randomly sets input values to zero to avoid overfitting (see the image below).

After calling the train method, the nolearn package will show status of the learning process, in my machine with my humble GPU I got the results below:

# Neural Network with 160362 learnable parameters

## Layer information

#  name      size
---  --------  --------
0  input     1x28x28
1  conv2d1   32x24x24
2  maxpool1  32x12x12
3  conv2d2   32x8x8
4  maxpool2  32x4x4
5  dropout1  32x4x4
6  dense     256
7  dropout2  256
8  output    10

epoch   train loss    valid loss    train/val    valid acc  dur
------- ------------  ------------  -----------  ---------  ---
1     0.85204   0.16707      5.09977      0.95174  33.71s
2     0.27571   0.10732      2.56896      0.96825  33.34s
3     0.20262   0.08567      2.36524      0.97488  33.51s
4     0.16551   0.07695      2.15081      0.97705  33.50s
5     0.14173   0.06803      2.08322      0.98061  34.38s
6     0.12519   0.06067      2.06352      0.98239  34.02s
7     0.11077   0.05532      2.00254      0.98427  33.78s
8     0.10497   0.05771      1.81898      0.98248  34.17s
9     0.09881   0.05159      1.91509      0.98407  33.80s
10     0.09264   0.04958      1.86864      0.98526  33.40s

As you can see, the accuracy in the end was 0.98526, a pretty good performance for a 10 epochs training.

### Prediction and Confusion Matrix

Now we can use the model to predict the entire testing dataset:

preds = net1.predict(X_test)

And we can also plot a confusion matrix to check the performance of the neural network classification:

cm = confusion_matrix(y_test, preds)
plt.matshow(cm)
plt.title('Confusion matrix')
plt.colorbar()
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.show()

The code above will plot the following confusion matrix:

As you can see, the diagonal is where the classification is more dense, showing the good performance of our classifier.

### Filters Visualization

We can also visualize the 32 filters from the first convolutional layer:

visualize.plot_conv_weights(net1.layers_['conv2d1'])

The code above will plot the following filters below:

As you can see, the nolearn plot_conv_weights plots all the filters present in the layer we specified.

### Theano layer functions and Feature Extraction

Now it is time to create theano-compiled functions that will feed-forward the input data into the architecture up to the layer you’re interested. I’m going to get the functions for the output layer and also for the dense layer before the output layer:

dense_layer = layers.get_output(net1.layers_['dense'], deterministic=True)
output_layer = layers.get_output(net1.layers_['output'], deterministic=True)
input_var = net1.layers_['input'].input_var

f_output = theano.function([input_var], output_layer)
f_dense = theano.function([input_var], dense_layer)

As you can see, we have now two theano functions called f_output and f_dense (for the output and dense layers). Please note that in order to get the layers here we are using a extra parameter called “deterministic“, this is to avoid the dropout layers affecting our feed-forward pass.

We can now convert an example instance to the input format and then feed it into the theano function for the output layer:

instance = X_test[0][None, :, :]
%timeit -n 500 f_output(instance)

500 loops, best of 3: 858 µs per loop

As you can see, the f_output function takes an average of 858 µs. We can also plot the output layer activations for the instance:

pred = f_output(instance)
N = pred.shape[1]
plt.bar(range(N), pred.ravel())

The code above will create the following plot:

As you can see, the digit was recognized as the digit 7. The fact that you can create theano functions for any layer of the network is very useful because you can create a function (like we did before) to get the activations for the dense layer (the one before the output layer) and you can use these activations as features and use your neural network not as classifier but as a feature extractor. Let’s plot now the 256 unit activations for the dense layer:

pred = f_dense(instance)
N = pred.shape[1]
plt.bar(range(N), pred.ravel())

The code above will create the following plot below:

You can now use the output of the these 256 activations as features on a linear classifier like Logistic Regression or SVM.

I hope you enjoyed the tutorial !

Cite this article as: Christian S. Perone, "Deep learning – Convolutional neural networks and feature extraction with Python," in Terra Incognita, 19/08/2015, https://blog.christianperone.com/2015/08/convolutional-neural-networks-and-feature-extraction-with-python/.

## Google’s S2, geometry on the sphere, cells and Hilbert curve

Update – 05 Dec 2017: Google just announced that it will be commited to the development of a new released version of the S2 library, amazing news, repository can be found here.

Google’s S2 library is a real treasure, not only due to its capabilities for spatial indexing but also because it is a library that was released more than 4 years ago and it didn’t get the attention it deserved. The S2 library is used by Google itself on Google Maps, MongoDB engine and also by Foursquare, but you’re not going to find any documentation or articles about the library anywhere except for a paper by Foursquare, a Google presentation and the source code comments. You’ll also struggle to find bindings for the library, the official repository has missing Swig files for the Python library and thanks to some forks we can have a partial binding for the Python language (I’m going to it use for this post). I heard that Google is actively working on the library right now and we are probably soon going to get more details about it when they release this work, but I decided to share some examples about the library and the reasons why I think that this library is so cool.

### The way to the cells

You’ll see this “cell” concept all around the S2 code. The cells are an hierarchical decomposition of the sphere (the Earth on our case, but you’re not limited to it) into compact representations of regions or points. Regions can also be approximated using these same cells, that have some nice features:

• They are compact (represented by 64-bit integers)
• They have resolution for geographical features
• They are hierarchical (thay have levels, and similar levels have similar areas)
• The containment query for arbitrary regions are really fast

The S2 library starts by projecting the points/regions of the sphere into a cube, and each face of the cube has a quad-tree where the sphere point is projected into. After that, some transformation occurs (for more details on why, see the Google presentation) and the space is discretized, after that the cells are enumerated on a Hilbert Curve, and this is why this library is so nice, the Hilbert curve is a space-filling curve that converts multiple dimensions into one dimension that has an special spatial feature: it preserves the locality.

### Hilbert Curve

The Hilbert curve is space-filling curve, which means that its range covers the entire n-dimensional space. To understand how this works, you can imagine a long string that is arranged on the space in a special way such that the string passes through each square of the space, thus filling the entire space. To convert a 2D point along to the Hilbert curve, you just need select the point on the string where the point is located. An easy way to understand it is to use this iterative example where you can click on any point of the curve and it will show where in the string the point is located and vice-versa.

In the image below, the point in the very beggining of the Hilbert curve (the string) is located also in the very beginning along curve (the curve is represented by a long string in the bottom of the image):

Now in the image below where we have more points, it is easy to see how the Hilbert curve is preserving the spatial locality. You can note that points closer to each other in the curve (in the 1D representation, the line in the bottom) are also closer in the 2D dimensional space (in the x,y plane). However, note that the opposite isn’t quite true because you can have 2D points that are close to each other in the x,y plane that aren’t close in the Hilbert curve.

Since S2 uses the Hilbert Curve to enumerate the cells, this means that cell values close in value are also spatially close to each other. When this idea is combined with the hierarchical decomposition, you have a very fast framework for indexing and for query operations. Before we start with the pratical examples, let’s see how the cells are represented in 64-bit integers.

If you are interested in Hilbert Curves, I really recommend this article, it is very intuitive and show some properties of the curve.

### The cell representation

As I already mentioned, the cells have different levels and different regions that they can cover. In the S2 library you’ll find 30 levels of hierachical decomposition. The cell level and the area range that they can cover is shown in the Google presentation in the slide that I’m reproducing below:

As you can see, a very cool result of the S2 geometry is that every cm² of the earth can be represented using a 64-bit integer.

The cells are represented using the following schema:

The first one is representing a leaf cell, a cell with the minimum area usually used to represent points. As you can see, the 3 initial bits are reserved to store the face of the cube where the point of the sphere was projected, then it is followed by the position of the cell in the Hilbert curve always followed by a “1” bit that is a marker that will identify the level of the cell.

So, to check the level of the cell, all that is required is to check where the last “1” bit is located in the cell representation. The checking of containment, to verify if a cell is contained in another cell, all you just have to do is to do a prefix comparison. These operations are really fast and they are possible only due to the Hilbert Curve enumeration and the hierarchical decomposition method used.

### Covering regions

So, if you want to generate cells to cover a region, you can use a method of the library where you specify the maximum number of the cells, the maximum cell level and the minimum cell level to be used and an algorithm will then approximate this region using the specified parameters. In the example below, I’m using the S2 library to extract some Machine Learning binary features using level 15 cells:

The cells regions are here represented in the image above using transparent polygons over the entire region of interest of my city. Since I used the level 15 both for minimum and maximum level, the cells are all covering similar region areas. If I change the minimum level to 8 (thus allowing the possibility of using larger cells), the algorithm will approximate the region in a way that it will provide the smaller number of cells and also trying to keep the approximation precise like in the example below:

As you can see, we have now a covering using larger cells in the center and to cope with the borders we have an approximation using smaller cells (also note the quad-trees).

### Examples

* In this tutorial I used the Python 2.7 bindings from the following repository. The instructions to compile and install it are present in the readme of the repository so I won’t repeat it here.

The first step to convert Latitude/Longitude points to the cell representation are shown below:

>>> import s2
>>> latlng = s2.S2LatLng.FromDegrees(-30.043800, -51.140220)
>>> cell = s2.S2CellId.FromLatLng(latlng)
>>> cell.level()
30
>>> cell.id()
10743750136202470315
>>> cell.ToToken()
951977d377e723ab

As you can see, we first create an object of the class S2LatLng to represent the lat/lng point and then we feed it into the S2CellId class to build the cell representation. After that, we can get the level and id of the class. There is also a method called ToToken that converts the integer representation to a compact alphanumerical representation that you can parse it later using FromToken method.

You can also get the parent cell of that cell (one level above it) and use containment methods to check if a cell is contained by another cell:

>>> parent = cell.parent()
>>> print parent.level()
29
>>> parent.id()
10743750136202470316
>>> parent.ToToken()
951977d377e723ac
>>> cell.contains(parent)
False
>>> parent.contains(cell)
True

As you can see, the level of the parent is one above the children cell (in our case, a leaf cell). The ids are also very similar except for the level of the cell and the containment checking is really fast (it is only checking the range of the children cells of the parent cell).

These cells can be stored on a database and they will perform quite well on a BTree index.  In order to create a collection of cells that will cover a region, you can use the S2RegionCoverer class like in the example below:

>>> region_rect = S2LatLngRect(
S2LatLng.FromDegrees(-51.264871, -30.241701),
S2LatLng.FromDegrees(-51.04618, -30.000003))
>>> coverer = S2RegionCoverer()
>>> coverer.set_min_level(8)
>>> coverer.set_max_level(15)
>>> coverer.set_max_cells(500)
>>> covering = coverer.GetCovering(region_rect)

First of all, we defined a S2LatLngRect which is a rectangle delimiting the region that we want to cover. There are also other classes that you can use (to build polygons for instance), the S2RegionCoverer works with classes that uses the S2Region class as base class. After defining the rectangle, we instantiate the S2RegionCoverer and then set the aforementioned min/max levels and the max number of the cells that we want the approximation to generate.

If you wish to plot the covering, you can use Cartopy, Shapely and matplotlib, like in the example below:

import matplotlib.pyplot as plt

from s2 import *

from shapely.geometry import Polygon

import cartopy.crs as ccrs
import cartopy.io.img_tiles as cimgt

proj = cimgt.MapQuestOSM()
plt.figure(figsize=(20,20), dpi=200)
ax = plt.axes(projection=proj.crs)

ax.set_extent([-51.411886, -50.922470,
-30.301314, -29.94364])

region_rect = S2LatLngRect(
S2LatLng.FromDegrees(-51.264871, -30.241701),
S2LatLng.FromDegrees(-51.04618, -30.000003))

coverer = S2RegionCoverer()
coverer.set_min_level(8)
coverer.set_max_level(15)
coverer.set_max_cells(500)
covering = coverer.GetCovering(region_rect)

geoms = []
for cellid in covering:
new_cell = S2Cell(cellid)
vertices = []
for i in xrange(0, 4):
vertex = new_cell.GetVertex(i)
latlng = S2LatLng(vertex)
vertices.append((latlng.lat().degrees(),
latlng.lng().degrees()))

geo = Polygon(vertices)
geoms.append(geo)

print "Total Geometries: {}".format(len(geoms))

edgecolor='black', alpha=0.4)
plt.show()

And the result will be the one below:

There are a lot of stuff in the S2 API, and I really recommend you to explore and read the source-code, it is really helpful. The S2 cells can be used for indexing and in key-value databases, it can be used on B Trees with really good efficiency and also even for Machine Learning purposes (which is my case), anyway, it is a very useful tool that you should keep in your toolbox. I hope you enjoyed this little tutorial !

– Christian S. Perone

## Arduino and OLED display to monitor Redis for fun and profit

I’m working on a new platform (hardware, firmware and software) to create “Stat Cubes“, which are tiny devices with OLED displays and wireless to monitor services or anything you want. While working on it I’ve made a little proof-of-concept using Arduino to monitor Redis server statistics. The Stat Cubes will be open-source in future but I’ve open-sourced the code of the PoC using Arduino and OLED to monitor the Redis server using a Python monitor that sends data from Redis server to the Arduino if someone is interested.

The main idea of Stat Cubes is that you will be able to leave the tiny cubes on your desk or even carry them with you. It will be a long road before I get the first version ready but if people show interest on it I’ll certainly try to speed up things.

Below you can see a video of the display working, you can also visit the repository for more screenshots, information and source-code both for the monitoring application and also for the Arduino code.

See more about the project in the Github repository.

### Screenshots

I hope you liked it !