texor 0.1.0

A ai library combining the best of TensorFlow and PyTorch

Texor - Comprehensive AI Framework

Texor is a comprehensive AI framework that combines the best features of TensorFlow and PyTorch. It provides a high-level API while maintaining flexibility and performance through a hybrid backend system.

Key Features

1. Hybrid Backend

• Leverage the power of both TensorFlow and PyTorch
• Seamlessly switch between backends
• Automatic optimization based on use case

2. Core API

from texor.core import Tensor

# Create tensors from various sources
x = Tensor([[1, 2], [3, 4]])  # From Python list
x = Tensor(numpy_array)        # From NumPy array
x = Tensor(tf_tensor)         # From TensorFlow tensor
x = Tensor(torch_tensor)      # From PyTorch tensor

# Access data in multiple formats
numpy_data = x.numpy()
tf_data = x.tensorflow()
torch_data = x.pytorch()

3. Neural Network Layers

from texor.nn import Sequential, Linear, Conv2D, MaxPool2D, ReLU, Dropout

model = Sequential([
    Conv2D(in_channels=1, out_channels=32, kernel_size=3),
    ReLU(),
    MaxPool2D(kernel_size=2),
    Conv2D(in_channels=32, out_channels=64, kernel_size=3),
    ReLU(),
    MaxPool2D(kernel_size=2),
    Linear(in_features=1600, out_features=10)
])

4. Optimizers

from texor.optim import SGD, Adam, RMSprop

# Create optimizer
optimizer = Adam(model.parameters(), lr=0.001)
optimizer = SGD(model.parameters(), lr=0.01, momentum=0.9)

5. Loss Functions

from texor.nn import MSELoss, CrossEntropyLoss, BCELoss

# Use loss functions
criterion = CrossEntropyLoss()
loss = criterion(predictions, targets)

Installation

pip install texor

Command Line Interface (CLI)

Texor provides a powerful CLI with intuitive features:

# View environment and setup information
texor info

# List available modules
texor list

# Search for specific modules
texor list resnet

# Check environment and dependencies
texor check

CLI Features:

• Color Output: Messages, warnings, and errors with clear color coding
• Progress Bars: Visual progress for long-running tasks
• Interactive Interface: User-friendly command line operations
• System Information: Detailed environment and configuration details

Basic Example

from texor.nn import Sequential, Linear, ReLU
from texor.core import Tensor
import numpy as np

# Create model
model = Sequential([
    Linear(input_size=784, output_size=256),
    ReLU(),
    Linear(input_size=256, output_size=10)
])

# Compile model
model.compile(
    optimizer='adam',
    loss='categorical_crossentropy'
)

# Create sample data
x = np.random.randn(100, 784)
y = np.random.randint(0, 10, size=(100,))
y = np.eye(10)[y]  # One-hot encode

# Train model
model.fit(
    x=Tensor(x),
    y=Tensor(y),
    epochs=10,
    batch_size=32
)

MNIST Example

See examples/mnist_example.py for a complete example of training a CNN on the MNIST dataset.

API Documentation

Core Module

• Tensor: Basic class for tensor operations
• zeros, ones, randn: Tensor creation functions
• from_numpy, from_tensorflow, from_pytorch: Conversions from other formats

Neural Network (nn) Module

• Layers: Linear, Conv2D, MaxPool2D, Dropout
• Activations: ReLU, Sigmoid, Tanh
• Loss Functions: MSELoss, CrossEntropyLoss, BCELoss
• Model: Sequential - Easy-to-use API for model building

Optimizers Module

• SGD: Stochastic Gradient Descent with momentum
• Adam: Adam optimizer
• RMSprop: RMSprop optimizer

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for more details.

License

MIT License - see the LICENSE file for details.

Language Support

For Vietnamese documentation, please see README_VN.md.