search

Lightning-Universe / lightning-bolts

Toolbox of models, callbacks, and datasets for AI/ML researchers.
https://lightning-bolts.readthedocs.io
Apache License 2.0
1.66k stars 321 forks source link

Add Fully Connected Network for Regression #393

Open acxz opened 3 years ago

acxz commented 3 years ago

🚀 Feature

Adding a fully connected (FC) multilayer perceptron (MLP) Network for Regression tasks.

Motivation

For simple regression tasks FC networks are commonly used. I saw that we have a LinearRegression model, but it would also be nice to have a full blown MLP network that people can just plug and chug for there supervised learning tasks.

Pitch

under pl_bolts/models/regression/ add mlp_regression.py, with the ability to specify the hidden layers and the hidden dims.

Alternatives

Usually people just roll their own solution for regression so it be a bit too specific to have an MLP network inside the pl_bolts repo, but I'm sure people looking for vector input to vector output regression problems would appreciate and solution that can be immediately used. Haven't seen an out of the box MLP solution elsewhere (maybe b/c it is kinda easy to implement on your own).

Additional context

github-actions[bot] commented 3 years ago

Hi! thanks for your contribution!, great first issue!

akihironitta commented 3 years ago

@acxz Thank you for your suggestion! It'd be nice to have a simple MLP regressor in Bolts. Once we have positive comments from other core members, would you be interested in submitting a PR?

acxz commented 3 years ago

Yeah I have been spinning one up today and can PR it once I finish/clean it up.

djbiega commented 2 years ago

Hi, I'd be interested in adding this in as an easy first issue if this is still desired. Could I get this assigned to me? @Borda @akihironitta