mgruben / TransferLearningYourself

An exploration of the mechanics of transfer learning in TensorFlow

Transfer Learning from Scratch

An exploration of the mechanics of transfer learning in TensorFlow.

Introduction

This repo explores transfer learning in TensorFlow with a model that we made ourselves, on the familiar MNIST dataset.

The motivation for this notebook is to recreate the results of Google's Transfer Learning tutorial, but with an arbitarily-defined network of our own, rather than Inception v3 or Mobilenet, as Google provides for in their retrain.py script.

Installation

The preferred method for starting this notebook is to create a new environment in conda:

$ conda create --name TransferLearningYourself
$ source activate TransferLearningYourself
$ conda install pip
$ pip install -r requirements.txt
$ jupyter notebook