wes-lewis/stg

Python/R library for feature selection in neural nets. ("Feature selection using Stochastic Gates", ICML 2020)

Feature Selection using Stochastic Gates (STG)

Project Page|Paper

Feature Selection using Stochastic Gates (STG) is a method for feature selection in neural network estimation problems. The new procedure is based on probabilistic relaxation of the l0 norm of features, or the count of the number of selected features. The proposed framework simultaneously learns either a nonlinear regression or classification function while selecting a small subset of features.

Top: Each stochastic gate z_d is drawn from the STG approximation of the Bernoulli distribution (shown as the blue histogram on the right). Specifically, z_d is obtained by applying the hard-sigmoid function to a mean-shifted Gaussian random variable. Bottom: The z_d stochastic gate is attached to the x_d input feature, where the trainable parameter µ_d controls the probability of the gate being active

Python

Installation

Installation with pip

To install with pip, run the following command:

pip install --user stg

Installation from GitHub

You can also clone the repository and install manually:

git clone 
cd stg/python
python setup.py install --user

Usage

Once you install the library, you can import STG to create a model instance:

from stg import STG
model = STG(task_type='regression',input_dim=X_train.shape[1], output_dim=1, hidden_dims=[500, 50, 10], activation='tanh', optimizer='SGD', learning_rate=0.1, batch_size=X_train.shape[0], feature_selection=True, sigma=0.5, lam=0.1, random_state=1, device="cpu") 

model.fit(X_train, y_train, nr_epochs=3000, valid_X=X_valid, valid_y=y_valid, print_interval=1000)
# Start training...

For more details, please see our Colab notebooks:

• Regression example
• Classification example
• Cox example

R

Installation

You first need to install the python package.

Installation from CRAN

Run the following command in your R console:

install.packages("Rstg")

Installation from Github

git clone git://github.com/runopti/stg.git
cd stg/python
python setup.py install --user
cd ../Rstg
R CMD INSTALL .

Usage

Please set the python path for reticulate to the python environment that you install the python stg package via this command in your R console or at the beginning of your R script.

reticulate::use_python("path_to_your_python_env_with_stg")

Then you can instantiate a trainer:

stg_trainer <- stg(task_type='regression', input_dim=100L, output_dim=1L, hidden_dims = c(500,50, 10), activation='tanh', optimizer='SGD', learning_rate=0.1, batch_size=100L, feature_selection=TRUE, sigma=0.5, lam=0.1, random_state=0.1)

You can then fit the model to data as follows:

# After preparing `X_train`, `y_train`, `X_valid`, and `y_valid'
stg_trainer$fit(X_train, y_train, nr_epochs=5000L, valid_X=X_valid, valid_y=y_valid, print_interval=1000L)

You can save your trained model

stg_trainer$save_checkpoint('r_test_model.pth')

and load the model

stg_trainer$load_checkpoint('r_test_mode.pth')

Acknowledgements and References

We thank Junchen Yang for his help to develop the R wrapper. Some of our codebase and its structure is inspired by https://github.com/vacancy/Jacinle.

If you find our library useful in your research, please consider citing us:

@incollection{icml2020_5085,
 author = {Yamada, Yutaro and Lindenbaum, Ofir and Negahban, Sahand and Kluger, Yuval},
 booktitle = {Proceedings of Machine Learning and Systems 2020},
 pages = {8952--8963},
 title = {Feature Selection using Stochastic Gates},
 year = {2020}
}