src.superphot_plus.model.data

Module Contents

Classes

PosteriorSamplesGroup	Holds data from multiple objects' posterior objects.
TrainData	Holds train and validation datasets.
TestData	Holds information about testing data.

class PosteriorSamplesGroup

Holds data from multiple objects’ posterior objects.

posterior_objects: List[superphot_plus.posterior_samples.PosteriorSamples]

use_redshift_info: bool | None = False

ignore_param_idxs: List[int] | None

random_seed: int | None

__post_init__()

__iter__()

oversample(fits_per_majority_lc=1)

Oversamples, drawing from posteriors of a certain fit. Assumes goal_per_class is the number of majority class if not set.

Returns:

Tuple containing oversampled features and labels.

Return type:

tuple of np.ndarray

oversample_smote()

Uses SMOTE to oversample data from rarer classes.

split(split_frac=0.1, split_indices=None, shuffle=True)

canonicalize_labels()

Convert labels to canon labels.

make_binary(target_label='SN Ia')

Convert labels to a binary classification problem.

make_fully_redshift_independent()

Experimental! We can convert our shape parameters to be FULLY z-independent by instead using: tau_rise/gamma, tau_rise/tau_fall, beta*tau_rise

(but its log scale for tau_rise, gamma, tau_fall so add/subtract instead)

We do everything relative to tau_rise because that’s the first shape param to be measured in real time!

class TrainData

Holds train and validation datasets.

train_dataset: torch.utils.data.TensorDataset

val_dataset: torch.utils.data.TensorDataset

__iter__()

class TestData

Holds information about testing data.

test_features: numpy.ndarray

test_classes: numpy.ndarray

test_names: numpy.ndarray

__post_init__()

Ensure everything is numpy arrays.

__iter__()