train

Train the model with given hparams (HPARAMS_PATH, ID) using data from DATA_DIR and finally save the metadata into OUT_DIR/RESULT_FNAME and the model itself into OUT_DIR/MODEL_FNAME.

The hparams is a look-up performed over col id bearing value ID (as in command line option --id) within the pandas.DataFrame read from the json file HPARAMS_PATH.

The data is loaded using phs.datasetFunctions.loadData with argument DATA_DIR, and return a tuple xTrainVal,yTrainVal,xTest,yTest. Data sanitisation is taken care of in the pre-process step.

phs.trainFunctions.Ksplit is responsible for shuffling and splitting the indices into train and val.

phs.trainFunctions.getTrainedModel is responsible for training the model. It returns a model with metadata of the form:

{ "model": "...<the python model>...", 
  "hparams": {"C": 0.7073982632},
  "metrics": {"valAcc": 0.8333333333333334},
  "theMetric": "valAcc"
}

The collation is based on the value of the key in dict metrics given by theMetric.

Usage:

train [OPTIONS]

Options:

  --out-dir DIRECTORY   [required]
  --result-fname TEXT   [required]
  --model-fname TEXT    [required]
  --hparams-path FILE   [required]
  --data-dir DIRECTORY  [required]
  --id TEXT             [required]
  --help                Show this message and exit.