Casper

some random functions/classes which serve a utilitarian purpose in multiple projects PS: Hierarchy of Classes will change as I add more utilities

Requirements

• pandas
• tensorboard
• pytorch

Installing

git clone https://github.com/chahuja/pycasper.git
cd pycasper
python setup.py install

ProtoTyping

Name (When finding unique names becomes a bigger task than the task itself)

• Based on a Namespace of arguments and its values, Name creates a unique name.
• If all arguments are too much for the length of the name, it is possible to provide a subset of those arguments
• Read more in the docfile of the class Name

BookKeeper

• Logging routine for deep learning

Results

Visualize results stored by BookKeeper. I would recommend using jupyter lab / notebook to see all the results in a pretty format

import pandas as pd
from from pycasper.results import walkthroughResults
df, df_all = walkthroughResults('save/', args_subset=['exp', 'cpk', 'model', 'window_hop'],
                  res_subset=['train', 'dev', 'test'], val_key='dev')
print(df.sort_values(by='dev'))

PS: Ideally, it would work best if you use BookKeeper to store the results. But, if you store results in a json format, with 'train', 'test', 'dev' as dictionary keys storing lists of loss/metrics across number of epochs and args for the experiment as a python Dictionary, this function would work.

Results filename: <name>_res.json
Args filename: <name>_args.args
Both of them are json files