CNN-TensorFlow-Tutorial

A simple CNN tutorial prepared for the SUNY Oswego TensorFlow workshop & talk.

Preparation

Please make sure you have python 2 or 3 installed. If not go to the python website for the installation. Please make sure you have the correct version of pip installed. Please go to the tensorflow website and install the appropriate version of tensorflow.

Datasets

We will be using the MNIST dataset, which is comprised of grayscale images of digits [0-9].

• http://yann.lecun.com/exdb/mnist/

Input Pipeline (ETL Process)

1. Extract
2. Transform
3. Load

Model Architecture (CNN)

• Convolution Layers
• Pooling Layers
• Dense Layer
• Flatten
• Dropout
• Logits

Model Evaluation

• Placeholders
• Train Operation
• Loss Operation
• Accuracy Operation

Extras

• Saving the Model for later use
• Adding Summaries
• Adding Regularization
• Confusion Matrix
• Model Loading
• Using Tensorboard

See the (Extra) Results here

• Extra Results

Slides

• pdf

A More Formal Introduction

• https://codelabs.developers.google.com/codelabs/cloud-tensorflow-mnist/#0

Other Resources

• https://www.tensorflow.org/performance/datasets_performance
• https://colah.github.io/posts/2014-07-Understanding-Convolutions/
• https://colah.github.io/posts/2014-07-Conv-Nets-Modular/
• My Blog Post