ML Python

About

Snippets

ML Operations

Production Protocol

Conventions

2D Matrix: capitalize variable names

  • Case: features

1D Vector: lowercase variable names

  • Case: labels

TensorFlow

Sequential Model Considerations

Train / Test Sets

Decrease the test size for smaller overall data sets. (~25%) otherwise aim for ~33%.

Batch Size

If not specified, the default batch size is 32. When training model with verbose, output displays batch number rather than each training feature.

Resources

Training Outline

Example workflow using the Iris classification problem.

Troubleshooting (TF1)

Uninitialized Variable

Errors with this notification could be a result of the global variable initializer not being run or being run before a variable is created.

init=tf.global_variables_initializer()