Hyperparameters

General parameters

Argument	Type	Help
–seed	int	random seed to be used
–use_cuda	bool	Use GPU

Data parameters

Argument	Type	Help
–data	bool	if True, launch data creation
–dump_dir	str	path where data and models are dumped
–norm	str	Feature normalization used in training/validation: None, perfilter, global, cosmo, cosmo_quantile
–redshift	str	Host redshift used in training/validation: zpho, zspe or None
–source_data	str	Data source: photometry or salt
–no_overwrite	bool	If True, overwrite preprocessed dir when creating database
–data_fraction	float	Fraction of data to use
–override_source_data	str	Change the source data (use saltfit or photometry)

Training parameters

Argument	Type	Help
–train_rnn	bool	Train RNN model
–train_rf	bool	Train RandomForest model
–monitor_interval	int	Validate every monitor_interval epochs–metrics

Validation Parameters

Argument	Type	Help
–validate_rnn	bool	Validate RNN model
–validate_rf	bool	Validate RandomForest model
–speed	bool	Run RNN speed classification benchmark
–calibration	bool	Evaluate model calibration
–performance	bool	Get performance metrics + plots
–metrics	bool	Compute performance metrics
–science_plots	bool	Plots of scientific interest
–model_files	bool	Path to model files
–prediction_files	bool	Path to prediction files
–metric_files	bool	Path to metric files

Visualization Parameters

Argument	Type	Help
–explore_lightcurves	bool	Plot a random selection of lightcurves
–plot_lcs	bool	Plot a random selection of lightcurves predictions
–plot_prediction_distribution	bool	Plot lcs and the histogram of probability for each class

RNN parameters

Argument	Type	Help
–cyclic	bool	Use cyclic learning rate
–cyclic_phases	list	Cyclic phases
–random_length	bool	Use random length sequences for training
–random_redshift	bool	If True, randomly set the spectroscopic redshift
–weight_decay	float	L2 decay on weights (for variational RNN)
–layer_type	str	Recurrent layer type. Choose lstm,gru,rnn
–model	str	Recurrent model type. Choose vanilla,variational,bayesian
–learning_rate	float	Learning rate
–nb_classes	int	Number of classification targets
–nb_epoch	int	Number of epoch
–batch_size	int	Batch size
–hidden_dim	int	Hidden layer dimension
–num_layers	int	Number of recurrent layers
–dropout	float	Dropout value
–bidirectional	bool	Use bidirectional models
–rnn_output_option	str	RNN output options. standard or mean
–pi	float	mixing coefficient for Bayes prior
–log_sigma1	float	Initialization parameter for BayesRNN layers
–log_sigma2	float	Initialization parameter for BayesRNN layers
–rho_scale_lower	float	Initialization parameter for BayesRNN layers
–rho_scale_upper	float	Initialization parameter for BayesRNN layers
–log_sigma1_output	float	Initialization parameter for BayesLinear output layers
–log_sigma2_output	float	Initialization parameter for BayesLinear output layers
–rho_scale_lower_output	float	Initialization parameter for BayesLinear output layers
–rho_scale_upper_output	float	Initialization parameter for BayesLinear output layers
–num_inference_samples	int	Number of samples to use for Bayesian inference
–mean_field_inference	bool	Use mean field inference for bayesian models

Random Forest parameters

Argument	Type	Help
–bootstrap	bool	Activate bootstrap when building trees
–min_samples_leaf	int	Minimum samples required to be a leaf node
–n_estimators	int	Number of trees
–min_samples_split	int	Min samples to create split
–criterion	str	Tree splitting criterion
–max_features	int	Max features per tree
–max_depth	int	Max tree depth