Neural network library based on theano
The goal of this project is to cover all the material of the official Theano deep learning tutorial and implement the appropriate classes and functions in Python 3.
numpy, scipy, theano for computations
scikit-learn for some helpful utilities
Setup your virtual environment as you like, navigate to a temp directory and execute:
git clone https://github.com/sotlampr/theano-wrapper
cd theano-wrapper
pip install requirements.txt
pip install -e ./
For a demo, open a python interpreter and type:
>>> from theano_wrapper.demo import demo
>>> demo()
For a complete documentation visit the read the docs page.
-
Regression estimators
- Linear Regression
- Multilayer Linear Regression
-
Classification estimators
- Logistic Regression
- Multilayer Perceptron
-
Unsupervised
- Single hidden layer tied autoencoder
- Denoising autoencoder
-
Training Classes
- Simple epoch-based gradient descent training
- (Minibatch) Stohastic gradient descent training
-
Regularization
- L1 and L2 squared
-
> 95% testing coverage
- More Estimators and trainers [coming soon]
- Documentation
- Error handling
- Testing for extreme cases
TODO