Keras model builder

What is this?

This is a tool build Keras models a lot faster than by coding. No coding is required, all can be done from a configuration file. The ModelBuilder not only builds a model from your configuration, but it also tries to make this process easier. Using analyzers your model is analyzed to prevent errors. For completeness, a serializer is also included to convert your model to a configuration file. There is also a framework present for hyperparameter optimalization, but this is currently limited to Hill Climbing.

Configuration parameters

Currently the code supports the following parameters:

• Model parameters
  • Loss
  • Activation
  • Optimizer
  • Metrics (optional)
  • Learning rate (optional)
• Dense layers
  • Units
  • Dropout (optional)
• Conv2D layers
  • Filters
  • Kernel size
  • Stride (optional)
  • Max Pooling (optional, can not be combined with Average Pooling)
  • Average Pooling (optional, can not be combined with Max Pooling)
  • Dropout (optional)
• RNN, GRU and LSTM layers
  • Units
  • Dropout (optional)
  • Bidirectional (optional)
• Gaussian Noise layers
  • Stddev

Why did you make this software?

Creating machine learning models in Keras is quite easy, but I think it can be easier. Editing layers, adding layers, removing layers, it all requires coding. Coding takes time, and code has to be debugged. That is why I created this tool, were the structure of a machine learning model is merely configuration, a set of variables in a file. It also makes it easier to create machine learning models from other code, and to use libraries different than Keras if wanted.

But Keras has a model format already

Yes, and you can do more with it than with this tool. The API of this tool is more important than the json format, which is still a reason to create this tool. The ModelBuilder also tries to help you, making the creation of models easier. Input, Activation and Reshape layers are automatically created where necessary.

Requirements

• Python >= 3.6
• Tensorflow