nn_models module

This module defines the Keras architectures for the Neural Networks used in the project. It includes simple models, multi-input models, and functions for hyperparameter tuning.

nn_models.final_model(flux_band_shape, flux_hist_shape, input_data_shape)[source]

Constructs the final three-input Neural Network architecture. Inputs are: Flux Band, Flux History, and Additional Features (e.g., Coordinates, Index).

Parameters:
  • flux_band_shape (tuple) – Shape tuple for Flux Band input.

  • flux_hist_shape (tuple) – Shape tuple for Flux History input.

  • input_data_shape (tuple) – Shape tuple for Additional Features input.

Returns:

A compiled Keras Model.

Return type:

keras.models.Model

nn_models.hp_final_model(hp)[source]

Constructs a tunable version of the final three-input architecture for Keras Tuner. Allows tuning of layer sizes, dropout, and learning rate.

Parameters:

hp (keras_tuner.HyperParameters) – Hyperparameters object from Keras Tuner.

Returns:

A compiled Keras Model ready for tuning.

Return type:

keras.models.Model

nn_models.hp_model_lr(hp)[source]

Builds a tunable sequential model that also tunes the learning rate.

Parameters:

hp (keras_tuner.HyperParameters) – Hyperparameters object from Keras Tuner.

Returns:

A compiled Keras Model with tunable learning rate.

Return type:

keras.models.Sequential

nn_models.paper_model(flux_band_shape, flux_hist_shape)[source]

Builds a multi-input neural network model based on the reference paper architecture. It takes flux bands and flux history as separate inputs.

Parameters:
  • flux_band_shape (tuple) – Shape of the flux band input features.

  • flux_hist_shape (tuple) – Shape of the flux history input features.

Returns:

A compiled Keras Model.

Return type:

keras.models.Model

nn_models.simple_model(input_data_shape)[source]

Builds a simple sequential Feed-Forward Neural Network.

Parameters:

input_data_shape (tuple) – Shape of the input features.

Returns:

A uncompiled Keras Sequential Model.

Return type:

keras.models.Sequential