def _run_quantiles_combiner_test(test: absltest.TestCase,
q_combiner: quantiles_util.QuantilesCombiner,
batches: List[List[np.ndarray]],
expected_result: np.ndarray):
"""Tests quantiles combiner."""
summaries = [q_combiner.add_input(q_combiner.create_accumulator(),
batch) for batch in batches]
result = q_combiner.extract_output(q_combiner.merge_accumulators(summaries))
test.assertEqual(result.dtype, expected_result.dtype)
test.assertEqual(result.size, expected_result.size)
for i in range(expected_result.size):
test.assertAlmostEqual(result[i], expected_result[i])
def _make_common_stats_proto(
common_stats: _PartialCommonStats,
parent_common_stats: Optional[_PartialCommonStats],
q_combiner: quantiles_util.QuantilesCombiner,
num_values_histogram_buckets: int,
has_weights: bool
) -> statistics_pb2.CommonStatistics:
"""Convert the partial common stats into a CommonStatistics proto."""
result = statistics_pb2.CommonStatistics()
result.num_non_missing = common_stats.num_non_missing
if parent_common_stats is not None:
result.num_missing = (
parent_common_stats.total_num_values - common_stats.num_non_missing)
result.tot_num_values = common_stats.total_num_values
# TODO(b/79685042): Need to decide on what is the expected values for
# statistics like min_num_values, max_num_values, avg_num_values, when
# all the values for the feature are missing.
if common_stats.num_non_missing > 0:
result.min_num_values = common_stats.min_num_values
result.max_num_values = common_stats.max_num_values
result.avg_num_values = (
common_stats.total_num_values / common_stats.num_non_missing)
# Add num_values_histogram to the common stats proto.
num_values_quantiles = q_combiner.extract_output(
common_stats.num_values_summary)
histogram = quantiles_util.generate_quantiles_histogram(
num_values_quantiles, common_stats.num_non_missing,
num_values_histogram_buckets)
result.num_values_histogram.CopyFrom(histogram)
# Add weighted common stats to the proto.
if has_weights:
weighted_common_stats_proto = statistics_pb2.WeightedCommonStatistics(
num_non_missing=common_stats.weighted_num_non_missing,
tot_num_values=common_stats.weighted_total_num_values)
if parent_common_stats is not None:
weighted_common_stats_proto.num_missing = (
parent_common_stats.weighted_total_num_values -
common_stats.weighted_num_non_missing)
if common_stats.weighted_num_non_missing > 0:
weighted_common_stats_proto.avg_num_values = (
common_stats.weighted_total_num_values /
common_stats.weighted_num_non_missing)
result.weighted_common_stats.CopyFrom(
weighted_common_stats_proto)
return result
def _make_numeric_stats_proto(
numeric_stats: _PartialNumericStats, total_num_values: int,
quantiles_combiner: quantiles_util.QuantilesCombiner,
num_histogram_buckets: int, num_quantiles_histogram_buckets: int,
has_weights: bool) -> statistics_pb2.NumericStatistics:
"""Convert the partial numeric statistics into NumericStatistics proto."""
result = statistics_pb2.NumericStatistics()
if numeric_stats.num_nan > 0:
total_num_values -= numeric_stats.num_nan
if total_num_values == 0:
# If we only have nan values, we only set num_nan.
if numeric_stats.num_nan > 0:
result.histograms.add(
type=statistics_pb2.Histogram.STANDARD).num_nan = (
numeric_stats.num_nan)
result.histograms.add(
type=statistics_pb2.Histogram.QUANTILES).num_nan = (
numeric_stats.num_nan)
return result
mean = numeric_stats.sum / total_num_values
variance = max(0, (numeric_stats.sum_of_squares / total_num_values) -
mean * mean)
result.mean = float(mean)
result.std_dev = math.sqrt(variance)
result.num_zeros = numeric_stats.num_zeros
result.min = float(numeric_stats.min)
result.max = float(numeric_stats.max)
# Extract the quantiles from the summary.
quantiles = quantiles_combiner.extract_output(
numeric_stats.quantiles_summary)
# Find the median from the quantiles and update the numeric stats proto.
result.median = float(quantiles_util.find_median(quantiles))
# Construct the equi-width histogram from the quantiles and add it to the
# numeric stats proto.
std_histogram = quantiles_util.generate_equi_width_histogram(
quantiles, numeric_stats.min, numeric_stats.max, total_num_values,
num_histogram_buckets)
std_histogram.num_nan = numeric_stats.num_nan
new_std_histogram = result.histograms.add()
new_std_histogram.CopyFrom(std_histogram)
# Construct the quantiles histogram from the quantiles and add it to the
# numeric stats proto.
q_histogram = quantiles_util.generate_quantiles_histogram(
quantiles, numeric_stats.min, numeric_stats.max, total_num_values,
num_quantiles_histogram_buckets)
q_histogram.num_nan = numeric_stats.num_nan
new_q_histogram = result.histograms.add()
new_q_histogram.CopyFrom(q_histogram)
# Add weighted numeric stats to the proto.
if has_weights:
weighted_numeric_stats_proto = statistics_pb2.WeightedNumericStatistics(
)
if numeric_stats.weighted_total_num_values == 0:
weighted_mean = 0
weighted_variance = 0
else:
weighted_mean = (numeric_stats.weighted_sum /
numeric_stats.weighted_total_num_values)
weighted_variance = max(0,
(numeric_stats.weighted_sum_of_squares /
numeric_stats.weighted_total_num_values) -
weighted_mean**2)
weighted_numeric_stats_proto.mean = weighted_mean
weighted_numeric_stats_proto.std_dev = math.sqrt(weighted_variance)
# Extract the weighted quantiles from the summary.
weighted_quantiles = quantiles_combiner.extract_output(
numeric_stats.weighted_quantiles_summary)
# Find the weighted median from the quantiles and update the proto.
weighted_numeric_stats_proto.median = float(
quantiles_util.find_median(weighted_quantiles))
# Construct the weighted equi-width histogram from the quantiles and
# add it to the numeric stats proto.
weighted_std_histogram = quantiles_util.generate_equi_width_histogram(
weighted_quantiles, numeric_stats.min, numeric_stats.max,
numeric_stats.weighted_total_num_values, num_histogram_buckets)
weighted_std_histogram.num_nan = numeric_stats.num_nan
weighted_numeric_stats_proto.histograms.extend(
[weighted_std_histogram])
# Construct the weighted quantiles histogram from the quantiles and
# add it to the numeric stats proto.
weighted_q_histogram = quantiles_util.generate_quantiles_histogram(
weighted_quantiles, numeric_stats.min, numeric_stats.max,
numeric_stats.weighted_total_num_values,
num_quantiles_histogram_buckets)
weighted_q_histogram.num_nan = numeric_stats.num_nan
weighted_numeric_stats_proto.histograms.extend([weighted_q_histogram])
result.weighted_numeric_stats.CopyFrom(weighted_numeric_stats_proto)
return result