TrainParticleClassifier#

class ctapipe.tools.train_particle_classifier.TrainParticleClassifier(**kwargs: Any)[source]#

Bases: Tool

Tool to train a ParticleClassifier on dl1b/dl2 data.

The tool first performs a cross validation to give an initial estimate on the quality of the estimation and then finally trains one model per telescope type on the full dataset.

Attributes Summary

`aliases`
`chunk_size`	How many subarray events to load at once before training on n_events (or all available) events.
`classes`
`description`
`examples`
`input_url_background`	Input dl1b/dl2 file for the background class.
`input_url_signal`	Input dl1b/dl2 file for the signal class.
`n_events`	Total number of events to be used for training.
`n_jobs`	Number of threads to use for the reconstruction.
`name`
`output_path`	Output file for the trained reconstructor.
`random_seed`	Random seed for sampling training events.
`signal_fraction`	Fraction of signal events in all events to be used for training.

Methods Summary

`finish`()	Write-out trained models and cross-validation results.
`setup`()	Initialize components from config.
`start`()	Train models per telescope type.

Attributes Documentation

aliases: StrDict = {'background': 'TrainParticleClassifier.input_url_background', 'cv-output': 'CrossValidator.output_path', 'n-events': 'TrainParticleClassifier.n_events', 'n-jobs': 'ParticleClassifier.n_jobs', 'signal': 'TrainParticleClassifier.input_url_signal', 'signal-fraction': 'TrainParticleClassifier.signal_fraction', ('o', 'output'): 'TrainParticleClassifier.output_path'}#

chunk_size#: How many subarray events to load at once before training on n_events (or all available) events.

classes: ClassesType = [<class 'ctapipe.io.tableloader.TableLoader'>, <class 'ctapipe.reco.sklearn.ParticleClassifier'>, <class 'ctapipe.reco.sklearn.CrossValidator'>]#

description: str | Unicode[str, str | bytes] = '\n Tool to train a `~ctapipe.reco.ParticleClassifier` on dl1b/dl2 data.\n\n The tool first performs a cross validation to give an initial estimate\n on the quality of the estimation and then finally trains one model\n per telescope type on the full dataset.\n '#

examples: str | Unicode[str, str | bytes] = '\n ctapipe-train-particle-classifier \\\n -c train_particle_classifier.yaml \\\n --signal gamma.dl2.h5 \\\n --background proton.dl2.h5 \\\n -o particle_classifier.pkl\n '#

input_url_background#: Input dl1b/dl2 file for the background class.

input_url_signal#: Input dl1b/dl2 file for the signal class.

n_events#: Total number of events to be used for training. If not given, all available events will be used (considering signal_fraction).

n_jobs#: Number of threads to use for the reconstruction. This overwrites the values in the config of each reconstructor.

name: str | Unicode[str, str | bytes] = 'ctapipe-train-particle-classifier'#

output_path#: Output file for the trained reconstructor. At the moment, pickle is the only supported format.

random_seed#: Random seed for sampling training events.

signal_fraction#: Fraction of signal events in all events to be used for training. signal_fraction = n_signal / (n_signal + n_background)

Methods Documentation

finish()[source]#: Write-out trained models and cross-validation results.

setup()[source]#: Initialize components from config.

start()[source]#: Train models per telescope type.