def test_can_upsample_bins_raises_if_target_bin_edges_do_not_line_up(self):
bins = [AgeBin(0,5), AgeBin(5, 10), AgeBin(10, 15)]
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 15)))
# a variety of misalignments relative to stated bins
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 14)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(5, 11)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(4, 10)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(1, 15)))
def test_can_upsample_bins_works_properly(self):
bins = [AgeBin(0,5), AgeBin(5, 10), AgeBin(10, 15)]
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 15)))
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 10)))
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(5, 10)))
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(5, 15)))
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 5)))
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(10, 15)))
def test_can_upample_bins_raises_if_target_bin_not_contained_by_bins(self):
bins = [AgeBin(0,5), AgeBin(5, 10), AgeBin(10, 15)]
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 16)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(-1, 15)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(15, 99)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(16, 99)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(-5, 0)))
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(-5, 1)))
def upsample_agebin(grouped_data, age_bin, aggregated_cols, weighted_cols, weighting_col):
"""
Upsample a pandas DataFrame object containing a AgeBin column to the requested age_bin. Intended to be supplied
as a dataframe groupby argument (to run on each group). It is ok for data outside the requested upsample range
to be in this dataframe; it will simply be excluded in the result.
Example usage: age_stratified_dataframe.groupby(['Year', 'Gender'].apply(upsample_agebin, AgeBin(15, 49))
:param grouped_data: a pandas DataFrameGroupBy object, see above.
:param age_bin: an AgeBin object representing inclusive lower and exclusive upper bounds.
:param weighted_cols: columns in the grouped data/dataframe to do weighted sums of
:return: A pandas DataFrame object with one row conataining the requested AgeBin-upsampled result
"""
# Further notes:
# verify we can do the requested upsample; this requires EXACT stitching of 'AgeBin' values to contain age_bin,
# though data outside the requested range will be ignored here and in the upsaampling.
if not AgeBin.can_upsample_bins(grouped_data['AgeBin'], age_bin):
raise NotUpsampleable('Cannot upsample to age bin: %s . Data is missing.' % age_bin)
# filter out data rows that are out of our requested age range, e.g. [50:55) is not in range of [15:49)
filtered_df = grouped_data.loc[[age_bin.contains(ab) for ab in grouped_data['AgeBin']]]
# grab row 0 and keep it as our base result; apply upsampled AgeBin
result = grouped_data[0:1].reset_index(drop=True)
result['AgeBin'] = str(age_bin)
# aggregated data items
total_weight = None
for channel in aggregated_cols:
total = np.sum(filtered_df[channel])
result[channel] = total
if channel == weighting_col: # hacky special case for use next
total_weight = total
# weighted sum items: model and reference data
fraction = filtered_df[weighting_col] / total_weight
for channel in weighted_cols:
result[channel] = np.sum(fraction * filtered_df[channel])
return result
def test_test_can_upsample_bins_works_with_AgeBin_and_string(self):
bins = [AgeBin(0, 5), '[5:::10)', AgeBin(10, 15)]
self.assertTrue(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(5, 15)))
def test_can_upsample_bins_works_if_no_bins_are_provided(self):
bins = []
self.assertFalse(AgeBin.can_upsample_bins(bins=bins, target_bin=AgeBin(0, 5)))