def test_rolling_similarity(self):
"""Should compute similarity between the sequence of vectors defined by the given columns, and given
similarity function"""
test_df = pd.DataFrame({"dx_dt": [1, 1, 1], "dy_dt": [1, 1, -1]})
transform_result = streams.rolling_similarity(test_df, cosine_similarity, "dx_dt", "dy_dt")
self.assertEqual(list(transform_result.columns.values), ["dx_dt", "dy_dt", "cosine_similarity_dx_dt_dy_dt"])
self.assertTrue(np.isnan(transform_result.cosine_similarity_dx_dt_dy_dt[0]))
self.assertAlmostEqual(transform_result.cosine_similarity_dx_dt_dy_dt[1], 1.0, 9)
self.assertAlmostEqual(transform_result.cosine_similarity_dx_dt_dy_dt[2], 0, 9)
def __total_deviation(self, activity_id):
stream_df = load_stream(self.database_driver, activity_id, 'latlng')
if stream_df is not None:
stream_df = lat_long_to_x_y(stream_df)
stream_df = smooth(stream_df, smooth_colname='x')
stream_df = smooth(stream_df, smooth_colname='y')
stream_df = derivative(stream_df, derivative_colname='x_smooth')
stream_df = derivative(stream_df, derivative_colname='y_smooth')
stream_df = rolling_similarity(stream_df, cosine_similarity, 'dx_smooth_dt', 'dy_smooth_dt')
try:
return np.nansum([cosine_to_deviation(cos)
for cos in stream_df.cosine_similarity_dx_smooth_dt_dy_smooth_dt])
except ValueError:
logging.warning('Failed to compute route deviation for activity %d, returning NaN' % activity_id)
return np.nan
else:
logging.warning('No gps stream available for activity %d, returning NaN' % activity_id)
return np.nan
