simple_function_test

Dependencies

• Python 3.6
• Tensorflow
• Keras
• Matplotlib

Description

• /networks
  • /dataset
    • dataset_generator.py generates data given a function object, splitting into training and testing
    • dataset_files.py save and load datasets from and to files
  • /models
    • initializers.py: Custom initializers for neural network model parameters using weight initialization procedures from: https://arxiv.org/abs/1704.08863
    • metrics.py: Mean sqaured error metric
    • model_base.py: Abstract Keras model class with training procedure
    • simple_dense.py: Keras model implementation of basic multi-layer dense network
  • /visualizers
    • simple_dense_visualizer.py: Visualization function for plotting data during training
• /simple_funcs
  • func_composition.py: Composes functions based on a single neural network layer into a single function object similar to a random dense neural network
  • function_base.py: Abstract base class for function objectives
  • step_funcs.py: Creates a function object for a step function of varying complexity, i.e. simple functions

• function_params.py: parameters used to create function object at runtime
• model_params.py: parameters used for building neural network model at runtime
• main.py: running a training session given a model and a dataset
• generate_test_data.py: generate dataset files from a function object
• generate_stats_plots.py: Generate statistics and plots from a training session
• generate_qsubs.py: Used for generating qsub commands for running many tests on the EPFL cluster
• run_from_qsub.py: Used by the qsub commands to run training on an EPFL cluster node