fink-anomaly-detection-model

Fink anomaly detection model

Здесь пока куча косяков, в обозримом будущем постараюсь их поправить

A set of modules for training models for finding anomalies in photometric data. There are currently two entry points via the console: fink_ad_model_train and get_anomaly_reactions.

fink_ad_model_train

The module trains the AADForest model using expert reactions from the C055ZJJ6N2AE channels in Slack and -1001898265997 in Telegram. It creates the following files:

• _g_means.csv and _r_means.csv -- averages over the training dataset;
• _reactions_g.csv and _reactions_r.csv -- training datasets for additional training of the AADForest algorithm, based on expert reactions from Slack and Telegram channels;
• forest_g_AAD.onnx -- model for _g filter
• forest_r_AAD.onnx -- model for _r filter

optional arguments:

--dataset_dir DATASET_DIR Input dir for dataset (default: './lc_features_20210617_photometry_corrected.parquet')

--n_jobs N_JOBS
Number of threads (default: -1)

usage: fink_ad_model_train [-h] [--dataset_dir DATASET_DIR] [--n_jobs N_JOBS]

get_anomaly_reactions

Uploading anomaly reactions from messengers. It creates the following files:

• _reactions_g.csv and _reactions_r.csv -- training datasets for additional training of the AADForest algorithm, based on expert reactions from Slack and Telegram channels;

optional arguments:

--slack_channel SLACK_CHANNEL Slack Channel ID (default: 'C055ZJJ6N2AE')

--tg_channel TG_CHANNEL Telegram Channel ID (default: -1001898265997)

usage: get_anomaly_reactions [-h] [--slack_channel SLACK_CHANNEL] [--tg_channel TG_CHANNEL]