dewball345/tfsim-benchmarking-test

Test benchmarking of Tensorflow Similarity using TFRecordDatasetSampler

TF similarity benchmarking but with TFRecordDatasetSampler.

1. Run save_tfrecords_classes.py to generate training, testing, seen queries, unseen queries, and index partitions. Each folder has tfrecords separated by class
2. Run train.py to train and save model
3. Run evaluate.py to get evaluation metrics

Problems - cannot see binary accuracy while training due to RAM limits

When evaluating cannot sample from unseen queries, seen queries, or index datasets because of not getting enough classes? Have just used some code from https://github.com/tensorflow/similarity/blob/kaggle/examples/supervised/kaggle_keras_tuner.ipynb which does this sampling. Cannot use all of test dataset due to ram limits as well.