Machine Learning

• scikit-learn

• tensorflow

orbithunter.machine_learning.orbit_cnn(orbits, target, **kwargs)

Create and train a deep learning model with 2 convolutional and 2 dense layers with Orbit state input Should be used as a crude reference due to its hard-coding.

Parameters

orbitsnumpy.ndarray of orbits.

Array of orbit states with shape (n_samples, time_discretization, space_discretization) shaped such that first axis is batch size or number of samples, then the last two dimensions are the ‘image’ dimensions, i.e. the two dimensions to convolve over. I.e. shape for KSE fields is

targetnumpy.ndarray

Must have same length along first axis as orbits. Can be any type of labels/values the dimension of each sample is the same as the dimension of the prediction/output layer.

kwargsdict, optional

May contain any and all extra keyword arguments required for numerical methods and Orbit specific methods.

hyper_parameters : tuple

Hyper parameters for deep learning layers.

Returns

tensorflow.keras.models.Sequential, tf.keras.callbacks.History, tuple

The model, its History (training and testing error as a function of epoch number) and tuple containing the train test splits of the regressors and target data. Train test split returned as (X_train, X_test, y_train, y_test).