FastNet (Not ready for install yet)

API layer built on tf-2.0 for high performance and easy training

Tenets

• We support only tensorflow 2.x, not 1.x support
• Emphasis on speed of training, when we have to choose between speed and code complexity, we will favor speed.
• Emphasis to build a high level API like Fast.ai on tensorflow
• Self contained but extendable.
• Every function should have proper documentation
• Notebook based examples should be provided
• Enable Multiple Runs and logging for seeing history of experiments.

Installation

pip install --upgrade --upgrade-strategy only-if-needed https://github.com/faizanahemad/FastNet/tarball/master

or for developer mode:

pip install -e .

Usage

Immediate Things to do

• setup unit test suite
• setup Site Page

References

Tensorflow Resources

• https://www.tensorflow.org/guide/performance/overview
• https://www.tensorflow.org/beta/guide/effective_tf2
• https://www.tensorflow.org/beta/guide/data_performance
• https://www.tensorflow.org/tutorials/load_data/tf_records
• https://github.com/kalaspuffar/tensorflow-data
• https://www.tensorflow.org/datasets/overview
• https://www.tensorflow.org/guide/performance/datasets
• https://www.tensorflow.org/beta/guide/keras/training_and_evaluation

Keras

• https://www.tensorflow.org/beta/guide/keras/overview

Others

• https://stackoverflow.com/questions/8161617/how-can-i-specify-library-versions-in-setup-py
• https://packaging.python.org/discussions/install-requires-vs-requirements/