/library

ML library, with prebuilt models, and useful functions and classes.

Primary LanguagePython

library

This is my Machine Learning Library! This library includes custom functionalities for:

  • machine learing models,
  • dataset loading
  • visualizations.

Check out examples/ for some example code on how to use this library.

To contribute, make sure that examples/ contains some example code on how to run your code.

utils/ contains all the functions and classes that can be used

Environment used and tested on:

  • This code is mainly made for Tensorflow 2
  • python 3.7

To use this library:

  • Add this library to your PYTHONPATH

utils/:

  • check out tf_custom for custom machine learning models and losses

dataset.py

To add a dataset with pre-existing loading functionalities:

  1. define your datapath. See utils/general_constants.py, overwrite datapath.
  2. Format the datasets

    • celeba:

      • make sure it is named celeba in your datapath and contains:
        • images
        • list_bbox_celeba.txt
        • identity_CelebA.txt
        • list_landmarks_align_celeba.txt
        • list_attr_celeba.txt
        • list_landmarks_celeba.txt

    • celeba-HQ:

      • make sure it is named celeba-hq in your datapath and contains:
        • celeba-1024 (contains only 1024x1024 images)
        • celeba-512 (contains only 512x512 images)
        • celeba-256 (contains only 256x256 images)
        • celeba-128 (contains only 128x128 images)
        • celeba-64 (contains only 64x64 images)
        • list_bbox_celeba.txt
        • list_landmarks_align_celeba.txt
        • list_landmarks_celeba.txt
        • identity_CelebA.txt
        • list_attr_celeba.txt
  3. to make these all into valid hdf5 datasets, run examples/dataset_test.py

For an example on how to use the dataset see examples/dataset_test.py

To add in a custom dataset:

  • put it into the datasets file.
  • For large datasets, use the GetData class which can retrieve images and labels and labels and save them into a .hdf5 file.

to use the built in celeba/celeba-hq datasets, see the comments in the get_celeba_dataset

If you want to load a dataset into a hdf5 file, you'll additionally need opencv. I only use openCV to quickly open an image to be loaded.

utils/tf_custom/:

Check out the different custom tensorflow models and functions here!