grid_feature_extractor.py

"""
Extract features from a set of spatially gridded data.

This feature extractor can be used for any problem in which one has a series of
spatial grids arranged in time, y, x dimensions. The xarray library is used
to read the data into a Dataset, which is represented by X_ds.
y_array can be a set of binary labels or a regressor value. The values in
y_array are assumed to be paired exactly with the values in the first
dimension of X_ds.
"""
import imp
import pandas as pd


class GridFeatureExtractor(object):
    def __init__(self, workflow_element_names=['feature_extractor']):
        self.element_names = workflow_element_names

    def train_submission(self, module_path, X_ds, y_array, train_is=None):
        if train_is is None:
            train_is = slice(None, None, None)
        submitted_feature_extractor_file = '{}/{}.py'.format(
            module_path, self.element_names[0])
        feature_extractor = imp.load_source(
            self.element_names[0], submitted_feature_extractor_file)
        fe = feature_extractor.FeatureExtractor()
        dim_set = pd.Series(list(X_ds.dims.keys()))
        time_dim = dim_set[dim_set.str.contains("time")][0]
        fe.fit(X_ds.isel(**{time_dim: train_is}), y_array[train_is])
        return fe

    def test_submission(self, trained_model, X_df):
        fe = trained_model
        X_test_array = fe.transform(X_df)
        return X_test_array