Trainbench

Trainbench is a tool for quickly setup a (Deep Learning) cross-validation training session with configuration instantiation, learned weights, plots, etc, recording at each training epoch.

Compatible with Keras.

⚠️ This package is currently under active development!

Getting started

Install

pip install trainbench

Create a train session

# train.py

# name
name = 'experiment_01'

# parameters to cross validate against
parameters = {
    'fc_size': [256, 512, 1024]    
    # ...
}

def train(parameters):
    xyz = parameters['xyz']
    # ...
    
    

Running from the command line

trainbench .

Checkout your results

    <name>/crosses/
                   /001 
                       ├─ meta.yml 
                       ├─ history.pkl 
                       └─ weights/
                              ├─ 001.h5
                              └─ 002.h5
                                 ...

In Jupyter Notebooks/Python script

from trainbench import Bench

parameters = {

}

def train_fn(parameters):
    pass
    
bench = Bench()
bench.train('experiment_xyz', train_fn, parameters)

Author notes

I'm wrote this tool for my own usage. Feel free to use it at your own will. If you would like to see any additional features / report existing issues please submit a pull request and/or open an issue.