scripts_description_en.md

Test Scripts Description

This document provides a brief overview of the scripts located in the test directory. These scripts are primarily used for testing and demonstrating the capabilities of the DeepFlows framework, including various model architectures and datasets.

CUDA Backend Tests

• test_cuda.py: Low-level testing script for the CUDA backend extension (CUDA_BACKEND). It verifies basic operations like array creation, memory transfer, and kernel execution (e.g., fill).

Basic Machine Learning Examples

• LinearRegression.py: A simple linear regression example using CPU. Fits a line $y = 3x + 5$ to synthetic data.

MLP (Multi-Layer Perceptron)

• MLP_MNIST.py: 3-layer MLP trained on MNIST dataset using CPU.
• MLP_MNIST_cuda.py: 3-layer MLP trained on MNIST dataset using CUDA acceleration.

CNN (Convolutional Neural Networks)

MNIST Dataset

• CNN_MNIST.py: CNN model trained on MNIST dataset using CPU.
• CNN_MNIST_cuda.py: CNN model trained on MNIST dataset using CUDA.

CIFAR-10 Dataset

• CNN_CIFAR10.py: CNN model trained on CIFAR-10 dataset (CPU).
• CNN_CIFAR10_cuda.py: CNN model trained on CIFAR-10 dataset (CUDA).
• CNN_CIFAR10_cuda_model_save_load_test.py: Tests the model saving and loading functionality for the CIFAR-10 CNN model on CUDA.

Animal-10 Dataset

• CNN_Animal10_cuda.py: CNN model trained on the Animal-10 dataset using CUDA.
• CNN_Animal10_cudacopy.py: Backup/Duplicate of the Animal-10 training script.

Dishes Dataset

• CNN_Dishes_cuda.py: CNN model trained on the Dishes dataset using CUDA. Adapted from the Animal-10 script.

Advanced Architectures

ResNet

• ResNet.py: Implementation of ResNet components (Residual Blocks).
• ResNet_Animal10_cuda.py: ResNet model trained on Animal-10 dataset using CUDA.
• ResNet_CIFAR10_cuda.py: ResNet model trained on CIFAR-10 dataset using CUDA.

MobileNet

• MobileNet.py: Implementation of MobileNetV1 architecture (Depthwise Separable Convolutions).

Pretrained Models & Transfer Learning

• test_pretrained_models.py: Tests the functionality of downloading, converting, and loading pretrained models. Demonstrates how to use the DeepFlows.utils.pretrained_models module for transfer learning.

---

Note: Scripts with _cuda suffix require a GPU and the DeepFlows CUDA backend to be properly built and configured.