esauidk/tensorflowfun

My Exploration With Tensorflow

TensorflowFun

My Exploration With Tensorflow!

Through out this repo there will be many different models being created.

Models.py was created to help with creation, testing, tuning, loading, and saving models.

Models.py

This script takes in command line inputs such as the model to use and the hyperparamters to be passed. With these inputs, the given model is executed.

Command Line Inputs

Here are the types of inputs to pass to this script:

• --modelName or -m (string) (REQURED) : Represents which model to excute (Can only choose models in the list of choices)
• --mode or -o (string) : Represents how to interact with the given mode (Current Choices are: Train, Test, Train&Test)
• --learningRate or -l (float) : Represents the learning rate to be passed to the model
• --batchSize or -b (int) : Represents the batch size to be passed to the model (the amount of training examples to work through before updating weights)
• --epochs or -e (int) : Represents the epochs to be passed to the model (the amount of full run throughs of the training data)
• --validationSplit or -v (float) : Represents the precentage of the test data to be used as a validation set
• --load-model (boolean flag) : Tells the model script to load a saved version of the model
• --save-model (boolean flag) : Tells the model script to save the model at the end of execution

Adding Models to Models.py

To add models to Models.py there are a couple of things to change to Models.py

Steps:

1. Add your model name to the MODEL_CHOICES list
2. Add your model function as a key-value entry in MODEL_FUNCTIONS. With the model name you added to MODEL_CHOICES as the key and the model's function as the value (Take a look at how the MNIST Neural Network Model is added)
3. In your model function, make sure to take a single arguement (call it argv if you'd like). This will hold the values of inputs passed into the command line. To access you can do argv.{input name} (Ex: argv.learningRate). The inputs will follow camelCase style.