Overview Pytorch CNN

A modular library that allows you to perform rapid mulitclass classification provided a dataset

Guide

Format your dataset like so:
- data/datasetname:
  - test
    - classname_1
      - sample_1.jpg / png
      - etc..
    - classname_2
      - sample_1.jpg
      - etc..
  - train
    - classname_1
      - sample_1.jpg / png
      - etc..
    - classname_2
      - sample_1.jpg
      - etc..
Run command python main.py train -t=image -m=<config_file_name (without the .yaml extension)>
To test model with cifar10 dataset run with command python main.py train -t=text -m=cifar_config dataset will automaticall download and run

Currently only supports using 1D CNNs with character level tokenization (ie: poor final testset accuracy)
Setup: Format dataset into csv file category column & text column symbol level tokenization is used as a default word level tokenization coming soon.
Then add a config.yaml file under configs (sample config provided under configs) and run command python main.py train -t='text' -m='<config_file_name> (without .yaml)'