Quick Start Guide - Ellipse Documentation

This guide assumes you have no prior knowledge of C, C++, PyTorch or any other language or deep learning framework.

First you have to download ellipse library from GitHub.

git clone https://github.com/oderoi/ellipse.git

Then create a working directory in your machine/Laptop folder, open ellipse and copy ellipse.h and paste it in your folder you created. Then create a new_file that you will use to code your project.

Then open the new_file in your favorite code editor and start to code.

Start by importing the library.

#include "ellipse.h"

Now you are set.

Tensor

Tensors are the base data structure in ellipse. They are like multi-dimensional array of a specific data type. All higher-level operations in ellipse operate on these tensors.

Tensor can be created from existing data.

Tensor * t1 = tensor((float[]){1, 2, 3, 4, 5, 6}, FLOAT32, (int[]){2, 3}, true);

So you might being asking yourself what the heck is that, don't sweat, let's see what the hell is that, and will start by looking one part after the other.

Tensor * t1: this is the variable definition where.
- Tensor *: telling our program that this is the Tensor type multidimensional array like.
- t1: is the name of our data, you can choose to use any name you like.
tensor(): is the function that helping us to hold together all the important information about out array, like data, data type, data dimension, data rank and requires grad.
- (float[]){1, 2, 3, 4, 5, 6}: this is the data itself that Tensor * t1 carries and (float[]) is the data type of the array, while the array itself is 1, 2, 3, 4, 5, 6.
- FLOAT32: this is the Tensor data type, which is supposed to be the same as (float[]).
- (int[]){2, 3}: is the array dimension. As well here (int[]) is the data type of the dimension array (and Yes, I just say array again, because array data and dimension are both represented using array) and the dimension itself is 2, 3.
- true: this boolean tell us that this Tensor will carry gradients of it's variable, during back propagation. So it might be true or false, depending on whether you want to calculate gradient correspond of the Tensor.

Tensor for different data types

int

Tensor *x = tensor((int[]){1,2,3,4}, INT, (int[]){2, 2}, false);

print(x);

Run

gcc nameOfFile.c -lm

./a.out

output

Tensor {
  dtype: int
  dims:  [2, 2]
  data:  [[1, 2],
          [3, 4]]
}

float32

Tensor *y = tensor((float[]){1,2,3,4,5,6}, FLOAT32, (int[]){3, 2}, true);

print(y);

Run

gcc nameOfFile.c -lm

./a.out

output

Tensor {
  dtype: float32
  dims:  [3, 2]
  data:  [[1.0000, 2.0000],
          [3.0000, 4.0000],
          [5.0000, 6.0000]]

  grads: [[0.0000e+00, 0.0000e+00],
          [0.0000e+00, 0.0000e+00],
          [0.0000e+00, 0.0000e+00]]
}

float64 / double

Tensor *x = tensor((double[]){1,2,3,4,5,6,7,8}, FLOAT64, (int[]){2, 4}, true);

print(x);

Run

gcc nameOfFile.c -lm

./a.out

output

Tensor {
  dtype: float64
  dims:  [2, 4]
  data:  [[1.0000, 2.0000, 3.0000, 4.0000],
          [5.0000, 6.0000, 7.0000, 8.0000]]

  grads: [[0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00],
          [0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00]]
}

← Documentation Home →