Machine Learning helpers#

ML helpers: synthetic dataset generation and simple Keras models.

This module provides small utilities to generate simulated datasets and a compact dense Keras model for parameter estimation. Note that some functions rely on external variables/functions (L, STAR_SIGNALS, kn_fields_jit, etc.). They are not executed during documentation builds.

phise.modules.ml.parameter_grid(N, D, a, b)[source]#

Generate a regular grid [a, b]^D with N steps per axis.

Parameters:

  • N – Resolution per axis (points per dimension).

  • D – Parameter space dimensionality.

  • a – Lower bound.

  • b – Upper bound.

Returns:

np.ndarray of shape (N**D, D) listing all grid points.

phise.modules.ml.parameter_basis(D, b=1)[source]#

Canonical basis augmented with the zero vector in R^D.

Parameters:

  • D – Dimension.

  • b – Norm of each basis vector (default 1).

Returns:

np.ndarray of shape (D+1, D).

phise.modules.ml.parameter_basis_2p(D, b=1)[source]#

Two-point basis: {0, b·e_i, 2b·e_i} for i=1..D.

Parameters:

  • D – Dimension.

  • b – Basis step.

Returns:

np.ndarray of shape (2D+1, D).

phise.modules.ml.get_dataset(size=1000, nb_shifters=14, nb_raw_outputs=7)[source]#

Build a structured synthetic dataset.

Note

Depends on external objects (e.g. L, STAR_SIGNALS, kn_fields_jit).

Parameters:

  • size – Number of samples to generate.

  • nb_shifters – Dimension of the output layer (number of parameters to predict).

  • nb_raw_outputs – Number of raw outputs representing the intensity mapping.

Returns:

np.ndarray of shape (size, vector_len).

phise.modules.ml.get_random_dataset(
size=1000,
nb_shifters=14,
nb_raw_outputs=7,
)[source]#

Build a random-point synthetic dataset.

Parameters:

  • size – Number of samples to generate.

  • nb_shifters – Expected number of shifters outputs.

  • nb_raw_outputs – Number of raw outputs.

Returns:

np.ndarray of shape (size, vector_len).

phise.modules.ml.get_model(input_shape, nb_shifters=14)[source]#

Create a small dense Keras network for output parameters.

Parameters:

  • input_shape – Input vector length.

  • nb_shifters – Expected number of outputs.

Returns:

Compiled tf.keras.Model (Adam optimizer, MSE loss).

phise.modules.ml.train_model(model, dataset, plot=True, nb_shifters=14)[source]#

Train the model and optionally plot the loss curves.

Parameters:

  • model – Compiled Keras model.

  • dataset – Data with targets in the last 14 columns.

  • plot – If True, plots loss and val_loss (log scale).

  • nb_shifters – Expected number of outputs.

Returns:

tf.keras.callbacks.History

phise.modules.ml.test_model(model, dataset, nb_shifters=14, nb_raw_outputs=7)[source]#

Plot ground-truth targets vs predictions for a quick visual check.

Parameters:

  • model – Trained Keras model.

  • dataset – Test data; uses 10 samples.

  • nb_shifters – Expected number of outputs.

  • nb_raw_outputs – expected number of raw outputs.

Returns:

None