#
# SPDX-FileCopyrightText: Copyright (c) 1993-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import copy
import functools
from collections import OrderedDict
from polygraphy import mod, util
from polygraphy.comparator import util as comp_util
from polygraphy.datatype import DataType
from polygraphy.logger import G_LOGGER, LogMode
np = mod.lazy_import("numpy")
[docs]@mod.export()
class OutputCompareResult:
"""
Represents the result of comparing a single output of a single iteration
between two runners.
"""
def __init__(
self,
passed,
max_absdiff,
max_reldiff,
mean_absdiff,
mean_reldiff,
median_absdiff,
median_reldiff,
quantile_absdiff,
quantile_reldiff,
):
"""
Records the required tolerances and other statistics gathered during comparison.
Args:
passed (bool):
Whether the error was within acceptable limits.
max_absdiff (float):
The minimum required absolute tolerance to consider the outputs equivalent.
max_reldiff (float):
The minimum required relative tolerance to consider the outputs equivalent.
mean_absdiff (float):
The mean absolute error between the outputs.
mean_reldiff (float):
The mean relative error between the outputs.
median_absdiff (float):
The median absolute error between the outputs.
median_reldiff (float):
The median relative error between the outputs.
quantile_absdiff (float):
The q-th quantile absolute error between the outputs.
quantile_reldiff (float):
The q-th quantile relative error between the outputs.
"""
self.passed = passed
self.max_absdiff = max_absdiff
self.max_reldiff = max_reldiff
self.mean_absdiff = mean_absdiff
self.mean_reldiff = mean_reldiff
self.median_absdiff = median_absdiff
self.median_reldiff = median_reldiff
self.quantile_absdiff = quantile_absdiff
self.quantile_reldiff = quantile_reldiff
[docs] def __bool__(self):
"""
Whether the output matched.
Returns:
bool
"""
return self.passed
def __str__(self):
return f"(atol={self.max_absdiff}, rtol={self.max_reldiff})"
[docs]@mod.export()
class DistanceMetricsResult:
"""
Represents the result of comparing a single output using distance metrics
between two runners.
"""
def __init__(
self,
passed,
l2_norm,
cosine_similarity,
):
"""
Records the distance metrics gathered during comparison.
Args:
passed (bool):
Whether the output passed all enabled metric comparisons.
l2_norm (float):
The L2 norm (Euclidean distance) between the outputs.
cosine_similarity (float):
The cosine similarity between the outputs.
"""
self.passed = passed
self.l2_norm = l2_norm
self.cosine_similarity = cosine_similarity
[docs] def __bool__(self):
"""
Whether the output passed all metric comparisons.
Returns:
bool
"""
return self.passed
[docs]@mod.export()
class QualityMetricsResult:
"""
Represents the result of comparing a single output using quality metrics
between two runners.
"""
def __init__(
self,
passed,
psnr=None,
snr=None,
):
"""
Records the quality metrics gathered during comparison.
Args:
passed (bool):
Whether the output passed all enabled quality metric comparisons.
psnr (float):
The Peak Signal-to-Noise Ratio between the outputs.
May be None if PSNR comparison was not enabled.
snr (float):
The Signal-to-Noise Ratio between the outputs.
May be None if SNR comparison was not enabled.
"""
self.passed = passed
self.psnr = psnr
self.snr = snr
[docs] def __bool__(self):
"""
Whether the output passed all metric comparisons.
Returns:
bool
"""
return self.passed
[docs]@mod.export()
class PerceptualMetricsResult:
"""
Represents the result of comparing a single output using perceptual metrics
between two runners.
"""
def __init__(
self,
passed,
lpips=None,
):
"""
Records the perceptual metrics gathered during comparison.
Args:
passed (bool):
Whether the output passed all enabled perceptual metric comparisons.
lpips (float):
The Learned Perceptual Image Patch Similarity score between the outputs.
Lower values indicate more perceptually similar outputs.
May be None if LPIPS computation failed.
"""
self.passed = passed
self.lpips = lpips
[docs] def __bool__(self):
"""
Whether the output passed all metric comparisons.
Returns:
bool
"""
return self.passed
def default_find_output_func(output_name, index, iter_result, base_iter_result):
found_name = util.find_str_in_iterable(output_name, iter_result.keys(), index)
if found_name is None:
return None
elif found_name != output_name:
exact_match = util.find_str_in_iterable(found_name, base_iter_result.keys())
if exact_match == found_name:
G_LOGGER.verbose(
f"Will not compare {found_name} with {output_name}, since the former already has an exact match: {exact_match}"
)
return None # If the found output is being compared against another output already, skip this non-exact match
G_LOGGER.warning(
f"Output names did not match exactly. Assuming {iter_result.runner_name} output: {found_name} corresponds to output: {output_name}"
)
return [found_name]
def run_comparison(func, fail_fast, iter_result0, iter_result1, find_output_func):
"""
Iterates over all the generated outputs and runs `func` to compare them.
"""
output_status = (
OrderedDict()
) # OrderedDict[str, bool] Maps output names to whether they matched.
for index, (out0_name, output0) in enumerate(iter_result0.items()):
out1_names = util.default(find_output_func(out0_name, index, iter_result1), [])
if len(out1_names) > 1:
G_LOGGER.info(
f"Will attempt to compare output: '{out0_name}' [{iter_result0.runner_name}] with multiple outputs: '{list(out1_names)}' [{iter_result1.runner_name}]"
)
for out1_name in out1_names:
if out1_name is None or out1_name not in iter_result1:
G_LOGGER.warning(
f"For output: '{out0_name}' [{iter_result0.runner_name}], skipping corresponding output: '{out1_name}' [{iter_result1.runner_name}], since the output was not found"
)
continue
output1 = iter_result1[out1_name]
G_LOGGER.start(
f"Comparing Output: '{out0_name}' (dtype={util.array.dtype(output0)}, shape={util.array.shape(output0)}) with '{out1_name}' (dtype={util.array.dtype(output1)}, shape={util.array.shape(output1)})"
)
with G_LOGGER.indent():
output_status[out0_name] = func(out0_name, output0, out1_name, output1)
if fail_fast and not output_status[out0_name]:
return output_status
mismatched_output_names = [
name for name, matched in output_status.items() if not matched
]
if mismatched_output_names:
G_LOGGER.error(f"FAILED | Mismatched outputs: {mismatched_output_names}")
else:
G_LOGGER.finish(
f"PASSED | All outputs matched | Outputs: {list(output_status.keys())}"
)
# This is useful for catching cases were Polygraphy does something wrong with the runner output buffers
if not output_status and (bool(iter_result0.keys()) or bool(iter_result1.keys())):
r0_name = iter_result0.runner_name
r0_outs = list(iter_result0.keys())
r1_name = iter_result1.runner_name
r1_outs = list(iter_result1.keys())
G_LOGGER.critical(
f"All outputs were skipped, no common outputs found! Note:\n{r0_name} outputs: {r0_outs}\n{r1_name} outputs: {r1_outs}"
)
return output_status
# Provides functions to compare two IterationResults
[docs]@mod.export()
class CompareFunc:
"""
Provides functions that can be used to compare two `IterationResult` s.
"""
[docs] @staticmethod
def simple(
check_shapes=None,
rtol=None,
atol=None,
fail_fast=None,
find_output_func=None,
check_error_stat=None,
infinities_compare_equal=None,
save_heatmaps=None,
show_heatmaps=None,
save_error_metrics_plot=None,
show_error_metrics_plot=None,
error_quantile=None,
):
"""
Creates a function that compares two IterationResults, and can be used as the `compare_func` argument
in ``Comparator.compare_accuracy``.
Args:
check_shapes (bool):
Whether shapes must match exactly. If this is False, this function may
permute or reshape outputs before comparison.
Defaults to True.
rtol (Union[float, Dict[str, float]]):
The relative tolerance to use when checking accuracy.
This is expressed as a percentage of the second set of output values.
For example, a value of 0.01 would check that the first set of outputs is within 1% of the second.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
Defaults to 1e-5.
atol (Union[float, Dict[str, float]]):
The absolute tolerance to use when checking accuracy.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
Defaults to 1e-5.
fail_fast (bool):
Whether the function should exit immediately after the first failure.
Defaults to False.
find_output_func (Callable(str, int, IterationResult) -> List[str]):
A callback that returns a list of output names to compare against from the provided
IterationResult, given an output name and index from another IterationResult.
The comparison function will always iterate over the output names of the
first IterationResult, expecting names from the second. A return value of
`[]` or `None` indicates that the output should be skipped.
check_error_stat (Union[str, Dict[str, str]]):
The error statistic to check. Possible values are:
- "elemwise": Checks each element in the output to determine if it exceeds both tolerances specified.
The minimum required tolerances displayed in this mode are only applicable when just one type of tolerance
is set. Because of the nature of the check, when both absolute/relative tolerance are specified, the required
minimum tolerances may be lower.
- "max": Checks the maximum absolute/relative errors against the respective tolerances. This is the strictest possible check.
- "mean" Checks the mean absolute/relative errors against the respective tolerances.
- "median": Checks the median absolute/relative errors against the respective tolerances.
- "quantile": Checks the quantile absolute/relative errors against the respective tolerances.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default error stat for outputs not explicitly listed.
Defaults to "elemwise".
infinities_compare_equal (bool):
If True, then matching +-inf values in the output have an absdiff of 0.
If False, then matching +-inf values in the output have an absdiff of NaN.
Defaults to False.
save_heatmaps (str):
[EXPERIMENTAL] Path to a directory in which to save figures of heatmaps of the absolute and relative error.
Defaults to None.
show_heatmaps (bool):
[EXPERIMENTAL] Whether to display heatmaps of the absolute and relative error.
Defaults to False.
save_error_metrics_plot (str):
[EXPERIMENTAL] Path to a directory in which to save the error metrics plots.
Defaults to None.
show_error_metrics_plot (bool):
[EXPERIMENTAL] Whether to display the error metrics plot.
error_quantile (Union[float, Dict[str, float]]):
Quantile error to compute when checking accuracy. This is expressed as a float in range [0, 1].
For example, error_quantile=0.5 is the median.
Defaults to 0.99.
Returns:
Callable(IterationResult, IterationResult) -> OrderedDict[str, OutputCompareResult]:
A callable that returns a mapping of output names to `OutputCompareResult` s, indicating
whether the corresponding output matched.
"""
check_shapes = util.default(check_shapes, True)
default_rtol = 1e-5
default_atol = 1e-5
default_quantile = 0.99
rtol = util.default(rtol, default_rtol)
atol = util.default(atol, default_atol)
error_quantile = util.default(error_quantile, default_quantile)
fail_fast = util.default(fail_fast, False)
default_error_stat = "elemwise"
check_error_stat = util.default(check_error_stat, default_error_stat)
infinities_compare_equal = util.default(infinities_compare_equal, False)
show_heatmaps = util.default(show_heatmaps, False)
show_error_metrics_plot = util.default(show_error_metrics_plot, False)
def check_outputs_match(
out0,
out0_name,
out1,
out1_name,
per_out_rtol,
per_out_atol,
per_out_err_stat,
runner0_name,
runner1_name,
per_out_quantile,
):
"""
Checks whether two outputs matched.
Args:
out0 (Union[np.array, torch.Tensor]): The first output.
out0_name (str): The name of the first output.
out1 (Union[np.array, torch.Tensor]): The second output.
out1_name (str): The name of the second output.
per_out_rtol (float): The relative tolerance to use for comparison.
per_out_atol (float): The absolute tolerance to use for comparison.
per_out_err_stat (str): The error statistic to check. See the docstring of ``simple`` for details.
runner0_name (str): The name of the runner that generated the first output.
runner1_name (str): The name of the runner that generated the second output.
per_out_quantile (float): The qunatile value to use for quantile comparison.
Returns:
OutputCompareResult: Details on whether the outputs matched.
"""
VALID_CHECK_ERROR_STATS = ["max", "mean", "median", "elemwise", "quantile"]
if per_out_err_stat not in VALID_CHECK_ERROR_STATS:
G_LOGGER.critical(
f"Invalid choice for check_error_stat: {per_out_err_stat}.\nNote: Valid choices are: {VALID_CHECK_ERROR_STATS}"
)
G_LOGGER.super_verbose(
f"{runner0_name:35} | Output: {out0_name} (dtype={util.array.dtype(out0)}, shape={util.array.shape(out0)}):\n{util.indent_block(out0)}"
)
G_LOGGER.super_verbose(
f"{runner1_name:35} | Output: {out1_name} (dtype={util.array.dtype(out1)}, shape={util.array.shape(out1)}):\n{util.indent_block(out1)}"
)
# Check difference vs. tolerances
if (
util.array.dtype(out0) == DataType.BOOL
and util.array.dtype(out1) == DataType.BOOL
):
absdiff = util.array.logical_xor(out0, out1)
else:
absdiff = util.array.abs(
util.array.subtract(
comp_util.cast_up(out0), comp_util.cast_up(out1)
)
)
if infinities_compare_equal:
out0_infinite = util.array.isinf(out0)
cond = util.array.logical_and(out0_infinite, out0 == out1)
absdiff = util.array.where(cond, 0, absdiff)
# Add a small epsilon (2e-16) to zero values in the array to prevent NaN in relative error.
out1_with_eps = copy.copy(comp_util.cast_up(out1))
if util.array.dtype(out1_with_eps).is_floating:
if util.array.any(out1_with_eps == 0):
G_LOGGER.warning(
f"{runner1_name:35} | Output: {out1_name}: Some values are 0. "
f"Will add a small epsilon quantity to these when computing relative difference. "
f"Note that this may cause some relative differences to be extremely high. ",
mode=LogMode.ONCE,
)
EPSILON = 2.220446049250313e-16
out1_with_eps[out1_with_eps == 0] += EPSILON
# TODO: Only evaluate this if actually needed like we do for quantile_*.
reldiff = util.array.divide(absdiff, util.array.abs(out1_with_eps))
min_reldiff = comp_util.compute_min(reldiff)
max_reldiff = comp_util.compute_max(reldiff)
mean_reldiff = comp_util.compute_mean(reldiff)
median_reldiff = comp_util.compute_median(reldiff)
quantile_reldiff = None
min_absdiff = comp_util.compute_min(absdiff)
max_absdiff = comp_util.compute_max(absdiff)
mean_absdiff = comp_util.compute_mean(absdiff)
median_absdiff = comp_util.compute_median(absdiff)
quantile_absdiff = None
def stat_failed(diff, tol):
return util.array.isnan(diff) or diff > tol
if per_out_err_stat == "mean":
failed = stat_failed(mean_absdiff, per_out_atol) and stat_failed(
mean_reldiff, per_out_rtol
)
elif per_out_err_stat == "median":
failed = stat_failed(median_absdiff, per_out_atol) and stat_failed(
median_reldiff, per_out_rtol
)
elif per_out_err_stat == "max":
failed = stat_failed(max_absdiff, per_out_atol) and stat_failed(
max_reldiff, per_out_rtol
)
elif per_out_err_stat == "quantile":
quantile_reldiff = comp_util.compute_quantile(reldiff, per_out_quantile)
quantile_absdiff = comp_util.compute_quantile(absdiff, per_out_quantile)
failed = stat_failed(quantile_absdiff, per_out_atol) and stat_failed(
quantile_reldiff, per_out_rtol
)
else:
assert (
per_out_err_stat == "elemwise"
), "This branch should be unreachable unless per_out_err_stat is 'elemwise'"
mismatches = (
util.array.greater(absdiff, per_out_atol)
| util.array.isnan(absdiff)
) & (
util.array.greater(reldiff, per_out_rtol)
| util.array.isnan(reldiff)
)
failed = util.array.any(mismatches)
try:
with G_LOGGER.indent():
G_LOGGER.super_verbose(
lambda: f"Mismatched indices:\n{util.array.argwhere(mismatches)}"
)
G_LOGGER.extra_verbose(
lambda: f"{runner0_name:35} | Mismatched values:\n{out0[mismatches]}"
)
G_LOGGER.extra_verbose(
lambda: f"{runner1_name:35} | Mismatched values:\n{out1[mismatches]}"
)
except Exception as err:
G_LOGGER.warning(
f"Failing to log mismatches.\nNote: Error was: {err}"
)
# Log information about the outputs
hist_bin_range = (
min(comp_util.compute_min(out0), comp_util.compute_min(out1)),
max(comp_util.compute_max(out0), comp_util.compute_max(out1)),
)
comp_util.log_output_stats(
out0, failed, f"{runner0_name}: {out0_name}", hist_range=hist_bin_range
)
comp_util.log_output_stats(
out1, failed, f"{runner1_name}: {out1_name}", hist_range=hist_bin_range
)
G_LOGGER.info(f"Error Metrics: {out0_name}")
with G_LOGGER.indent():
def req_tol(mean_diff, median_diff, max_diff, quantile_diff):
return {
"mean": mean_diff,
"median": median_diff,
"max": max_diff,
"elemwise": max_diff,
"quantile": quantile_diff,
}[per_out_err_stat]
msg = f"Minimum Required Tolerance: {per_out_err_stat} error | [abs={req_tol(mean_absdiff, median_absdiff, max_absdiff, quantile_absdiff):.5g}] OR [rel={req_tol(mean_reldiff, median_reldiff, max_reldiff, quantile_reldiff):.5g}]"
if per_out_err_stat == "elemwise":
msg += " (requirements may be lower if both abs/rel tolerances are set)"
elif per_out_err_stat == "quantile":
msg += f" (quantile={per_out_quantile:.4g})"
G_LOGGER.info(msg)
if save_error_metrics_plot or show_error_metrics_plot:
with G_LOGGER.indent():
comp_util.scatter_plot_error_magnitude(
absdiff,
reldiff,
comp_util.cast_up(out1),
min_reldiff,
max_reldiff,
runner0_name,
runner1_name,
out0_name,
out1_name,
save_dir=save_error_metrics_plot,
show=show_error_metrics_plot,
)
def build_heatmaps(diff, min_diff, max_diff, prefix, use_lognorm=None):
if save_heatmaps or show_heatmaps:
with G_LOGGER.indent():
comp_util.build_heatmaps(
diff,
min_diff,
max_diff,
prefix=f"{prefix} Error | {out0_name}",
save_dir=save_heatmaps,
show=show_heatmaps,
use_lognorm=use_lognorm,
)
comp_util.log_output_stats(absdiff, failed, "Absolute Difference")
build_heatmaps(absdiff, min_absdiff, max_absdiff, "Absolute")
comp_util.log_output_stats(reldiff, failed, "Relative Difference")
build_heatmaps(
reldiff, min_reldiff, max_reldiff, "Relative", use_lognorm=True
)
G_LOGGER.extra_verbose(
lambda: f"Finished comparing: '{out0_name}' (dtype={util.array.dtype(out0)}, shape={util.array.shape(out0)}) [{runner0_name}] and '{out1_name}' (dtype={util.array.dtype(out1)}, shape={util.array.shape(out1)}) [{runner1_name}]"
)
return OutputCompareResult(
not failed,
max_absdiff,
max_reldiff,
mean_absdiff,
mean_reldiff,
median_absdiff,
median_reldiff,
quantile_absdiff,
quantile_reldiff,
)
def compare_output(iter_result0, iter_result1):
"""
Compare the outputs of two runners from a single iteration.
This function will always iterate over the output names of the first IterationResult,
and attempt to find corresponding output names in the second.
If no corresponding output name is found, the output is skipped.
If all output names are skipped, then this function raises an error.
Args:
iter_result0 (IterationResult): The result of the first runner.
iter_result1 (IterationResult): The result of the second runner.
Returns:
OrderedDict[str, OutputCompareResult]:
The name of the outputs compared, derived from the first IterationResult,
and whether they matched. If an output name is not found, it is omitted from this dictionary.
Raises:
PolygraphyException: If all output names are skipped, and thus no outputs are compared.
"""
def check_dict(dct, dict_name):
if isinstance(dct, dict):
util.check_sequence_contains(
dct.keys(),
set(iter_result0.keys()) | set(iter_result1.keys()) | {""},
name=dict_name,
log_func=G_LOGGER.warning,
check_missing=False,
)
check_dict(rtol, "the rtol dictionary")
check_dict(atol, "the atol dictionary")
check_dict(check_error_stat, "the check_error_stat dictionary")
check_dict(error_quantile, "the quantile dictionary")
if not check_shapes:
G_LOGGER.info(
"Strict shape checking disabled. Will attempt to match output shapes before comparisons"
)
def match(out0_name, output0, out1_name, output1):
per_out_atol = util.value_or_from_dict(atol, out0_name, default_atol)
per_out_rtol = util.value_or_from_dict(rtol, out0_name, default_rtol)
per_out_err_stat = util.value_or_from_dict(
check_error_stat, out0_name, default_error_stat
)
per_out_quantile = util.value_or_from_dict(
error_quantile, out0_name, default_quantile
)
G_LOGGER.info(
f"Tolerance: [abs={per_out_atol:.5g}, rel={per_out_rtol:.5g}] | Checking {per_out_err_stat} error"
)
G_LOGGER.extra_verbose(
f"Note: Comparing {iter_result0.runner_name} vs. {iter_result1.runner_name}"
)
if check_shapes and util.array.shape(output0) != util.array.shape(
output1
):
G_LOGGER.error(
f"FAILED | Output: `{out0_name}` | Will not compare outputs of different shapes.\n"
f"Note: Output shapes are {util.array.shape(output0)} and {util.array.shape(output1)}."
)
G_LOGGER.error(
"Note: Use --no-shape-check or set check_shapes=False to "
"attempt to compare values anyway.",
mode=LogMode.ONCE,
)
return False
output1 = util.try_match_shape(output1, util.array.shape(output0))
output0 = util.array.view(
output0,
DataType.from_dtype(util.array.dtype(output0)),
util.array.shape(output1),
)
outputs_matched = check_outputs_match(
output0,
out0_name,
output1,
out1_name,
per_out_rtol=per_out_rtol,
per_out_atol=per_out_atol,
per_out_err_stat=per_out_err_stat,
runner0_name=iter_result0.runner_name,
runner1_name=iter_result1.runner_name,
per_out_quantile=per_out_quantile,
)
# Finally show summary.
if not outputs_matched:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | Difference exceeds tolerance (rel={per_out_rtol}, abs={per_out_atol})"
)
else:
G_LOGGER.finish(
f"PASSED | Output: '{out0_name}' | Difference is within tolerance (rel={per_out_rtol}, abs={per_out_atol})"
)
return outputs_matched
nonlocal find_output_func
find_output_func = util.default(
find_output_func,
functools.partial(
default_find_output_func, base_iter_result=iter_result0
),
)
return run_comparison(
match, fail_fast, iter_result0, iter_result1, find_output_func
)
return compare_output
[docs] @staticmethod
def indices(index_tolerance=None, fail_fast=None):
"""
Creates a function that compares two IterationResults containing indices, and can be used as the `compare_func` argument
in ``Comparator.compare_accuracy``. This can be useful to compare, for example, the outputs of a Top-K operation.
Outputs with more than one dimension are treated like multiple batches of values. For example, an output of shape (3, 4, 5, 10)
would be treated like 60 batches (3 x 4 x 5) of 10 values each.
Args:
index_tolerance (Union[int, Dict[str, int]]):
The tolerance to use when comparing indices. This is an integer indicating the maximum distance
between values before it is considered a mismatch. For example, consider two outputs:
::
output0 = [0, 1, 2]
output1 = [1, 0, 2]
With an index tolerance of 0, this would be considered a mismatch, since the positions of `0` and `1`
are flipped between the two outputs. However, with an index tolerance of 1, it would pass since
the mismatched values are only 1 spot apart. If instead the outputs were:
::
output0 = [0, 1, 2]
output1 = [1, 2, 0]
Then we would require an index tolerance of 2, since the `0` value in the two outputs is 2 spots apart.
When this value is set, the final 'index_tolerance' number of values are ignored for each batch.
For example, with an index tolerance of 1, mismatches in the final element are not considered.
If used with a Top-K output, you can compensate for this by instead using a Top-(K + index_tolerance).
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
fail_fast (bool):
Whether the function should exit immediately after the first failure.
Defaults to False.
Returns:
Callable(IterationResult, IterationResult) -> OrderedDict[str, bool]:
A callable that returns a mapping of output names to `bool` s, indicating
whether the corresponding output matched.
"""
index_tolerance = util.default(index_tolerance, 0)
fail_fast = util.default(fail_fast, False)
def compare_output(iter_result0, iter_result1):
"""
Compare the outputs of two runners from a single iteration.
This function will always iterate over the output names of the first IterationResult,
and attempt to find corresponding output names in the second.
If no corresponding output name is found, the output is skipped.
If all output names are skipped, then this function raises an error.
Args:
iter_result0 (IterationResult): The result of the first runner.
iter_result1 (IterationResult): The result of the second runner.
Returns:
OrderedDict[str, bool]:
The name of the outputs compared, derived from the first IterationResult,
and whether they matched. If an output name is not found, it is omitted from this dictionary.
Raises:
PolygraphyException: If all output names are skipped, and thus no outputs are compared.
"""
def match(out0_name, output0, out1_name, output1):
per_out_index_tol = util.value_or_from_dict(
index_tolerance, out0_name, 0
)
if util.array.shape(output0) != util.array.shape(output1):
G_LOGGER.error("Cannot compare outputs of different shapes.")
return False
passed = True
for batch in np.ndindex(util.array.shape(output0)[:-1]):
out0_vals = output0[batch]
if per_out_index_tol > 0:
out0_vals = out0_vals[:-per_out_index_tol]
out1_vals = output1[batch]
for index0, val0 in enumerate(out0_vals):
if val0 == out1_vals[index0]:
continue
index1 = util.array.ravel(
util.array.argwhere(out1_vals == val0)
)
if util.array.size(index1) < 1:
G_LOGGER.error(
f"FAILED | Value: {val0} not found in output"
)
passed = False
if fail_fast:
return False
continue
index1 = index1[0]
if abs(index1 - index0) > per_out_index_tol:
G_LOGGER.error(
f"FAILED | Difference exceeds index tolerance ({per_out_index_tol})"
)
passed = False
if fail_fast:
return False
continue
# Log information about the outputs
hist_bin_range = (
min(comp_util.compute_min(output0), comp_util.compute_min(output1)),
max(comp_util.compute_max(output0), comp_util.compute_max(output1)),
)
comp_util.log_output_stats(
output0,
not passed,
f"{iter_result0.runner_name}: {out0_name}",
hist_range=hist_bin_range,
)
comp_util.log_output_stats(
output1,
not passed,
f"{iter_result1.runner_name}: {out1_name}",
hist_range=hist_bin_range,
)
if passed:
G_LOGGER.finish(
f"PASSED | Difference is within index tolerance ({per_out_index_tol})"
)
return passed
return run_comparison(
match,
fail_fast,
iter_result0,
iter_result1,
functools.partial(
default_find_output_func, base_iter_result=iter_result0
),
)
return compare_output
[docs] @staticmethod
def distance_metrics(
l2_tolerance=None,
cosine_similarity_threshold=None,
check_shapes=None,
fail_fast=None,
find_output_func=None,
):
"""
Creates a function that compares two IterationResults using distance metrics (L2 norm and cosine similarity),
and can be used as the `compare_func` argument in ``Comparator.compare_accuracy``.
Args:
l2_tolerance (Union[float, Dict[str, float]]):
The tolerance to use when checking L2 norm (Euclidean distance).
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
Defaults to 1e-5.
cosine_similarity_threshold (Union[float, Dict[str, float]]):
The minimum cosine similarity required for outputs to be considered matching.
Cosine similarity measures the cosine of the angle between two vectors, with values between -1 and 1.
A value of 1 means vectors are identical or parallel, 0 means they are orthogonal, and -1 means they point in opposite directions.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default threshold for outputs not explicitly listed.
Defaults to 0.997 (which corresponds to a cosine distance of 0.003).
check_shapes (bool):
Whether shapes must match exactly. If this is False, this function may
permute or reshape outputs before comparison.
Defaults to True.
fail_fast (bool):
Whether the function should exit immediately after the first failure.
Defaults to False.
find_output_func (Callable(str, int, IterationResult) -> List[str]):
A callback that returns a list of output names to compare against from the provided
IterationResult, given an output name and index from another IterationResult.
The comparison function will always iterate over the output names of the
first IterationResult, expecting names from the second. A return value of
`[]` or `None` indicates that the output should be skipped.
Returns:
Callable(IterationResult, IterationResult) -> OrderedDict[str, DistanceMetricsResult]:
A callable that returns a mapping of output names to `DistanceMetricsResult` s, indicating
whether the corresponding output matched based on the distance metrics.
"""
check_shapes = util.default(check_shapes, True)
default_l2_tolerance = 1e-5
default_cosine_similarity_threshold = 0.997
l2_tolerance = util.default(l2_tolerance, default_l2_tolerance)
cosine_similarity_threshold = util.default(cosine_similarity_threshold, default_cosine_similarity_threshold)
fail_fast = util.default(fail_fast, False)
def compute_l2_norm(array1, array2):
"""Compute L2 norm (Euclidean distance) between two arrays."""
diff = util.array.subtract(comp_util.cast_up(array1), comp_util.cast_up(array2))
squared_diff = util.array.power(diff, 2)
sum_squared_diff = util.array.sum(squared_diff)
return util.array.sqrt(sum_squared_diff)
def compute_cosine_similarity(array1, array2):
"""Compute cosine similarity between two arrays."""
array1_flat = util.array.ravel(comp_util.cast_up(array1))
array2_flat = util.array.ravel(comp_util.cast_up(array2))
# Calculate dot product
dot_product = util.array.sum(util.array.multiply(array1_flat, array2_flat))
# Calculate magnitudes
magnitude1 = util.array.sqrt(util.array.sum(util.array.power(array1_flat, 2)))
magnitude2 = util.array.sqrt(util.array.sum(util.array.power(array2_flat, 2)))
# Avoid division by zero
if magnitude1 == 0 and magnitude2 == 0:
return 1.0 # If both vectors are zero, they are identical (similarity = 1.0)
elif magnitude1 == 0 or magnitude2 == 0:
return 0.0 # If only one vector is zero, they are orthogonal (similarity = 0.0)
# Cosine similarity is dot_product / (magnitude1 * magnitude2)
cosine_similarity = dot_product / (magnitude1 * magnitude2)
# Handle floating point issues that might make cosine_similarity slightly outside [-1, 1]
return max(-1.0, min(1.0, cosine_similarity))
def check_outputs_match(
out0,
out0_name,
out1,
out1_name,
per_out_l2_tol,
per_out_cosine_sim_threshold,
runner0_name,
runner1_name,
):
"""
Checks whether two outputs matched using L2 norm and cosine similarity.
Args:
out0 (Union[np.array, torch.Tensor]): The first output.
out0_name (str): The name of the first output.
out1 (Union[np.array, torch.Tensor]): The second output.
out1_name (str): The name of the second output.
per_out_l2_tol (float): The L2 norm tolerance to use for comparison.
per_out_cosine_sim_threshold (float): The minimum cosine similarity required for a match.
runner0_name (str): The name of the runner that generated the first output.
runner1_name (str): The name of the runner that generated the second output.
Returns:
DistanceMetricsResult: Details on whether the outputs matched.
"""
G_LOGGER.super_verbose(
f"{runner0_name:35} | Output: {out0_name} (dtype={util.array.dtype(out0)}, shape={util.array.shape(out0)}):\n{util.indent_block(out0)}"
)
G_LOGGER.super_verbose(
f"{runner1_name:35} | Output: {out1_name} (dtype={util.array.dtype(out1)}, shape={util.array.shape(out1)}):\n{util.indent_block(out1)}"
)
# Compute metrics
l2_norm = compute_l2_norm(out0, out1)
cosine_sim = compute_cosine_similarity(out0, out1)
# Check if outputs match based on the metrics
l2_passed = bool(l2_norm <= per_out_l2_tol)
cosine_passed = bool(cosine_sim >= per_out_cosine_sim_threshold)
# Overall pass requires all enabled metrics to pass
passed = bool(l2_passed and cosine_passed)
# Log information
hist_bin_range = (
min(comp_util.compute_min(out0), comp_util.compute_min(out1)),
max(comp_util.compute_max(out0), comp_util.compute_max(out1)),
)
comp_util.log_output_stats(
out0, not passed, f"{runner0_name}: {out0_name}", hist_range=hist_bin_range
)
comp_util.log_output_stats(
out1, not passed, f"{runner1_name}: {out1_name}", hist_range=hist_bin_range
)
G_LOGGER.info(f"Distance Metrics: {out0_name}")
with G_LOGGER.indent():
G_LOGGER.info(f"L2 Norm: {l2_norm:.5g} (tolerance: {per_out_l2_tol:.5g}) | {'PASSED' if l2_passed else 'FAILED'}")
G_LOGGER.info(f"Cosine Similarity: {cosine_sim:.5g} (threshold: {per_out_cosine_sim_threshold:.5g}) | {'PASSED' if cosine_passed else 'FAILED'}")
# Create a proper DistanceMetricsResult object with our metrics
result = DistanceMetricsResult(
passed=passed,
l2_norm=l2_norm,
cosine_similarity=cosine_sim
)
if not passed:
if not l2_passed:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | L2 Norm ({l2_norm:.5g}) exceeds tolerance ({per_out_l2_tol:.5g})"
)
if not cosine_passed:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | Cosine Similarity ({cosine_sim:.5g}) below threshold ({per_out_cosine_sim_threshold:.5g})"
)
else:
metrics_passed = ["L2 Norm", "Cosine Similarity"]
G_LOGGER.finish(
f"PASSED | Output: '{out0_name}' | All metrics passed: {', '.join(metrics_passed)}"
)
return result
def compare_output(iter_result0, iter_result1):
"""
Compare the outputs of two runners from a single iteration using distance metrics.
This function will always iterate over the output names of the first IterationResult,
and attempt to find corresponding output names in the second.
If no corresponding output name is found, the output is skipped.
If all output names are skipped, then this function raises an error.
Args:
iter_result0 (IterationResult): The result of the first runner.
iter_result1 (IterationResult): The result of the second runner.
Returns:
OrderedDict[str, DistanceMetricsResult]:
The name of the outputs compared, derived from the first IterationResult,
and whether they matched. If an output name is not found, it is omitted from this dictionary.
Raises:
PolygraphyException: If all output names are skipped, and thus no outputs are compared.
"""
def check_dict(dct, dict_name):
if isinstance(dct, dict):
util.check_sequence_contains(
dct.keys(),
set(iter_result0.keys()) | set(iter_result1.keys()) | {""},
name=dict_name,
log_func=G_LOGGER.warning,
check_missing=False,
)
check_dict(l2_tolerance, "the l2_tolerance dictionary")
check_dict(cosine_similarity_threshold, "the cosine_similarity_threshold dictionary")
if not check_shapes:
G_LOGGER.info(
"Strict shape checking disabled. Will attempt to match output shapes before comparisons"
)
def match(out0_name, output0, out1_name, output1):
per_out_l2_tol = util.value_or_from_dict(l2_tolerance, out0_name, default_l2_tolerance)
per_out_cosine_sim_threshold = util.value_or_from_dict(cosine_similarity_threshold, out0_name, default_cosine_similarity_threshold)
# Build tolerance message showing all enabled metrics
tolerance_msg = [f"L2={per_out_l2_tol:.5g}", f"Cosine Similarity min={per_out_cosine_sim_threshold:.5g}"]
G_LOGGER.info(f"Tolerance: [{', '.join(tolerance_msg)}]")
G_LOGGER.extra_verbose(
f"Note: Comparing {iter_result0.runner_name} vs. {iter_result1.runner_name}"
)
if check_shapes and util.array.shape(output0) != util.array.shape(
output1
):
G_LOGGER.error(
f"FAILED | Output: `{out0_name}` | Will not compare outputs of different shapes.\n"
f"Note: Output shapes are {util.array.shape(output0)} and {util.array.shape(output1)}."
)
G_LOGGER.error(
"Note: Use --no-shape-check or set check_shapes=False to "
"attempt to compare values anyway.",
mode=LogMode.ONCE,
)
return False
output1 = util.try_match_shape(output1, util.array.shape(output0))
output0 = util.array.view(
output0,
DataType.from_dtype(util.array.dtype(output0)),
util.array.shape(output1),
)
outputs_matched = check_outputs_match(
output0,
out0_name,
output1,
out1_name,
per_out_l2_tol=per_out_l2_tol,
per_out_cosine_sim_threshold=per_out_cosine_sim_threshold,
runner0_name=iter_result0.runner_name,
runner1_name=iter_result1.runner_name,
)
return outputs_matched
nonlocal find_output_func
find_output_func = util.default(
find_output_func,
functools.partial(
default_find_output_func, base_iter_result=iter_result0
),
)
return run_comparison(
match, fail_fast, iter_result0, iter_result1, find_output_func
)
return compare_output
[docs] @staticmethod
def quality_metrics(
psnr_tolerance=None,
snr_tolerance=None,
check_shapes=None,
fail_fast=None,
find_output_func=None,
):
"""
Creates a function that compares two IterationResults using quality metrics (PSNR and SNR),
and can be used as the `compare_func` argument in ``Comparator.compare_accuracy``.
Args:
psnr_tolerance (Union[float, Dict[str, float]]):
The minimum PSNR (Peak Signal-to-Noise Ratio) value required for outputs to be considered matching.
Higher values of PSNR indicate better quality matches. Typical acceptable values are 30 dB or above.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
If None, PSNR check will be skipped. Defaults to 30.0.
snr_tolerance (Union[float, Dict[str, float]]):
The minimum SNR (Signal-to-Noise Ratio) value required for outputs to be considered matching.
Higher values of SNR indicate better quality matches.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default tolerance for outputs not explicitly listed.
If None, SNR check will be skipped. Defaults to 20.0.
check_shapes (bool):
Whether shapes must match exactly. If this is False, this function may
permute or reshape outputs before comparison.
Defaults to True.
fail_fast (bool):
Whether the function should exit immediately after the first failure.
Defaults to False.
find_output_func (Callable(str, int, IterationResult) -> List[str]):
A callback that returns a list of output names to compare against from the provided
IterationResult, given an output name and index from another IterationResult.
The comparison function will always iterate over the output names of the
first IterationResult, expecting names from the second. A return value of
`[]` or `None` indicates that the output should be skipped.
Returns:
Callable(IterationResult, IterationResult) -> OrderedDict[str, QualityMetricsResult]:
A callable that returns a mapping of output names to `QualityMetricsResult` s, indicating
whether the corresponding output matched based on the quality metrics.
"""
check_shapes = util.default(check_shapes, True)
default_psnr_tolerance = 30.0
default_snr_tolerance = 20.0
psnr_tolerance = util.default(psnr_tolerance, default_psnr_tolerance)
snr_tolerance = util.default(snr_tolerance, default_snr_tolerance)
fail_fast = util.default(fail_fast, False)
def compute_psnr(array1, array2):
"""
Compute Peak Signal-to-Noise Ratio between two arrays.
Higher values indicate better matches.
"""
array1_cast = comp_util.cast_up(array1)
array2_cast = comp_util.cast_up(array2)
# Compute Mean Squared Error
mse = util.array.mean(util.array.power(
util.array.subtract(array1_cast, array2_cast), 2
))
# Avoid division by zero
if mse == 0:
return float('inf') # Perfect match
# Compute data range (max value in reference array)
max_val = comp_util.compute_max(array1_cast)
if max_val <= 0:
max_val = 1.0 # Default to 1.0 if max value is non-positive
# PSNR formula: 20 * log10(MAX) - 10 * log10(MSE)
psnr = 20 * np.log10(max_val) - 10 * np.log10(mse)
return psnr
def compute_snr(array1, array2):
"""
Compute Signal-to-Noise Ratio between two arrays.
Higher values indicate better matches.
"""
array1_cast = comp_util.cast_up(array1)
array2_cast = comp_util.cast_up(array2)
# Signal power
signal_power = util.array.mean(util.array.power(array1_cast, 2))
# Noise is the difference between the arrays
noise = util.array.subtract(array1_cast, array2_cast)
noise_power = util.array.mean(util.array.power(noise, 2))
# Avoid division by zero
if noise_power == 0:
return float('inf') # Perfect match
if signal_power == 0:
return -float('inf') # No signal
# SNR formula: 10 * log10(signal_power / noise_power)
snr = 10 * np.log10(signal_power / noise_power)
return snr
def check_outputs_match(
out0,
out0_name,
out1,
out1_name,
per_out_psnr_tol,
per_out_snr_tol,
runner0_name,
runner1_name,
):
"""
Checks whether two outputs matched using quality metrics (PSNR and SNR).
Args:
out0 (Union[np.array, torch.Tensor]): The first output.
out0_name (str): The name of the first output.
out1 (Union[np.array, torch.Tensor]): The second output.
out1_name (str): The name of the second output.
per_out_psnr_tol (float): The minimum PSNR value required for a match.
per_out_snr_tol (float): The minimum SNR value required for a match.
runner0_name (str): The name of the runner that generated the first output.
runner1_name (str): The name of the runner that generated the second output.
Returns:
QualityMetricsResult: Details on whether the outputs matched.
"""
G_LOGGER.super_verbose(
f"{runner0_name:35} | Output: {out0_name} (dtype={util.array.dtype(out0)}, shape={util.array.shape(out0)}):\n{util.indent_block(out0)}"
)
G_LOGGER.super_verbose(
f"{runner1_name:35} | Output: {out1_name} (dtype={util.array.dtype(out1)}, shape={util.array.shape(out1)}):\n{util.indent_block(out1)}"
)
# Compute metrics
psnr_value = None
if per_out_psnr_tol is not None:
psnr_value = compute_psnr(out0, out1)
snr_value = None
if per_out_snr_tol is not None:
snr_value = compute_snr(out0, out1)
# Check if outputs match based on the metrics
# Default to True for metrics that weren't computed
psnr_passed = True
if per_out_psnr_tol is not None and psnr_value is not None:
psnr_passed = bool(psnr_value >= per_out_psnr_tol)
snr_passed = True
if per_out_snr_tol is not None and snr_value is not None:
snr_passed = bool(snr_value >= per_out_snr_tol)
# Overall pass requires all enabled metrics to pass
passed = bool(psnr_passed and snr_passed)
# Log information
hist_bin_range = (
min(comp_util.compute_min(out0), comp_util.compute_min(out1)),
max(comp_util.compute_max(out0), comp_util.compute_max(out1)),
)
comp_util.log_output_stats(
out0, not passed, f"{runner0_name}: {out0_name}", hist_range=hist_bin_range
)
comp_util.log_output_stats(
out1, not passed, f"{runner1_name}: {out1_name}", hist_range=hist_bin_range
)
G_LOGGER.info(f"Quality Metrics: {out0_name}")
with G_LOGGER.indent():
if per_out_psnr_tol is not None and psnr_value is not None:
G_LOGGER.info(f"PSNR: {psnr_value:.5g} dB (min required: {per_out_psnr_tol:.5g} dB) | {'PASSED' if psnr_passed else 'FAILED'}")
if per_out_snr_tol is not None and snr_value is not None:
G_LOGGER.info(f"SNR: {snr_value:.5g} dB (min required: {per_out_snr_tol:.5g} dB) | {'PASSED' if snr_passed else 'FAILED'}")
# Create a proper QualityMetricsResult object with our metrics
result = QualityMetricsResult(
passed=passed,
psnr=psnr_value,
snr=snr_value
)
if not passed:
if per_out_psnr_tol is not None and psnr_value is not None and not psnr_passed:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | PSNR ({psnr_value:.5g} dB) below required minimum ({per_out_psnr_tol:.5g} dB)"
)
if per_out_snr_tol is not None and snr_value is not None and not snr_passed:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | SNR ({snr_value:.5g} dB) below required minimum ({per_out_snr_tol:.5g} dB)"
)
else:
metrics_passed = []
if per_out_psnr_tol is not None and psnr_value is not None:
metrics_passed.append("PSNR")
if per_out_snr_tol is not None and snr_value is not None:
metrics_passed.append("SNR")
if metrics_passed:
G_LOGGER.finish(
f"PASSED | Output: '{out0_name}' | All quality metrics passed: {', '.join(metrics_passed)}"
)
else:
G_LOGGER.warning(
f"PASSED | Output: '{out0_name}' | No quality metrics were successfully computed"
)
return result
def compare_output(iter_result0, iter_result1):
"""
Compare the outputs of two runners from a single iteration using quality metrics.
This function will always iterate over the output names of the first IterationResult,
and attempt to find corresponding output names in the second.
If no corresponding output name is found, the output is skipped.
If all output names are skipped, then this function raises an error.
Args:
iter_result0 (IterationResult): The result of the first runner.
iter_result1 (IterationResult): The result of the second runner.
Returns:
OrderedDict[str, QualityMetricsResult]:
The name of the outputs compared, derived from the first IterationResult,
and whether they matched. If an output name is not found, it is omitted from this dictionary.
Raises:
PolygraphyException: If all output names are skipped, and thus no outputs are compared.
"""
def check_dict(dct, dict_name):
if isinstance(dct, dict):
util.check_sequence_contains(
dct.keys(),
set(iter_result0.keys()) | set(iter_result1.keys()) | {""},
name=dict_name,
log_func=G_LOGGER.warning,
check_missing=False,
)
check_dict(psnr_tolerance, "the psnr_tolerance dictionary")
check_dict(snr_tolerance, "the snr_tolerance dictionary")
if not check_shapes:
G_LOGGER.info(
"Strict shape checking disabled. Will attempt to match output shapes before comparisons"
)
def match(out0_name, output0, out1_name, output1):
per_out_psnr_tol = None
if psnr_tolerance is not None:
per_out_psnr_tol = util.value_or_from_dict(psnr_tolerance, out0_name, default_psnr_tolerance)
per_out_snr_tol = None
if snr_tolerance is not None:
per_out_snr_tol = util.value_or_from_dict(snr_tolerance, out0_name, default_snr_tolerance)
# Build tolerance message showing all enabled metrics
tolerance_msg = []
if per_out_psnr_tol is not None:
tolerance_msg.append(f"PSNR min={per_out_psnr_tol:.5g} dB")
if per_out_snr_tol is not None:
tolerance_msg.append(f"SNR min={per_out_snr_tol:.5g} dB")
if tolerance_msg:
G_LOGGER.info(f"Quality Metrics Tolerance: [{', '.join(tolerance_msg)}]")
else:
G_LOGGER.warning("No quality metrics enabled for comparison")
G_LOGGER.extra_verbose(
f"Note: Comparing {iter_result0.runner_name} vs. {iter_result1.runner_name}"
)
if check_shapes and util.array.shape(output0) != util.array.shape(
output1
):
G_LOGGER.error(
f"FAILED | Output: `{out0_name}` | Will not compare outputs of different shapes.\n"
f"Note: Output shapes are {util.array.shape(output0)} and {util.array.shape(output1)}."
)
G_LOGGER.error(
"Note: Use --no-shape-check or set check_shapes=False to "
"attempt to compare values anyway.",
mode=LogMode.ONCE,
)
return False
output1 = util.try_match_shape(output1, util.array.shape(output0))
output0 = util.array.view(
output0,
DataType.from_dtype(util.array.dtype(output0)),
util.array.shape(output1),
)
outputs_matched = check_outputs_match(
output0,
out0_name,
output1,
out1_name,
per_out_psnr_tol=per_out_psnr_tol,
per_out_snr_tol=per_out_snr_tol,
runner0_name=iter_result0.runner_name,
runner1_name=iter_result1.runner_name,
)
return outputs_matched
nonlocal find_output_func
find_output_func = util.default(
find_output_func,
functools.partial(
default_find_output_func, base_iter_result=iter_result0
),
)
return run_comparison(
match, fail_fast, iter_result0, iter_result1, find_output_func
)
return compare_output
[docs] @staticmethod
def perceptual_metrics(
lpips_threshold=None,
check_shapes=None,
fail_fast=None,
find_output_func=None,
):
"""
Creates a function that compares two IterationResults using perceptual metrics (LPIPS),
and can be used as the `compare_func` argument in ``Comparator.compare_accuracy``.
This function specifically targets image-like data and uses perceptual similarity metrics
that correlate better with human perception than traditional distance metrics.
Args:
lpips_threshold (Union[float, Dict[str, float]]):
The maximum LPIPS (Learned Perceptual Image Patch Similarity) score allowed for outputs to be considered matching.
Lower values indicate more perceptually similar outputs. Typical values are below 0.1.
This can be provided on a per-output basis using a dictionary. In that case,
use an empty string ("") as the key to specify default threshold for outputs not explicitly listed.
If None, a default value of 0.1 will be used.
check_shapes (bool):
Whether shapes must match exactly. If this is False, this function may
permute or reshape outputs before comparison.
Defaults to True.
fail_fast (bool):
Whether the function should exit immediately after the first failure.
Defaults to False.
find_output_func (Callable(str, int, IterationResult) -> List[str]):
A callback that returns a list of output names to compare against from the provided
IterationResult, given an output name and index from another IterationResult.
The comparison function will always iterate over the output names of the
first IterationResult, expecting names from the second. A return value of
`[]` or `None` indicates that the output should be skipped.
Returns:
Callable(IterationResult, IterationResult) -> OrderedDict[str, PerceptualMetricsResult]:
A callable that returns a mapping of output names to `PerceptualMetricsResult` s, indicating
whether the corresponding output matched based on the perceptual metrics.
"""
check_shapes = util.default(check_shapes, True)
default_lpips_threshold = 0.1
lpips_threshold = util.default(lpips_threshold, default_lpips_threshold)
fail_fast = util.default(fail_fast, False)
# Try to import torch and lpips if available
torch = None
lpips_model = None
try:
torch = mod.lazy_import("torch")
lpips = mod.lazy_import("lpips")
# Initialize LPIPS model with explicit device specification
device = torch.device('cpu')
# Try with different initialization approaches
try:
# First try with default initialization
lpips_model = lpips.LPIPS(net='alex', version='0.1').to(device)
except Exception as e1:
G_LOGGER.warning(f"First LPIPS initialization approach failed: {e1}. Trying alternative method...")
try:
# Try with a different network if AlexNet fails
lpips_model = lpips.LPIPS(net='vgg', version='0.1').to(device)
except Exception as e2:
G_LOGGER.warning(f"Second LPIPS initialization approach failed: {e2}. Trying basic initialization...")
try:
# As a last resort, try with the most basic initialization
model = lpips.LPIPS(net_type='alex')
model.eval()
lpips_model = model.to(device)
except Exception as e3:
G_LOGGER.warning(
f"Failed to initialize LPIPS model with all methods. LPIPS check will be skipped. "
f"Errors: {e1}; {e2}; {e3}"
)
except ImportError:
G_LOGGER.warning(
"LPIPS comparison requested but torch or lpips module not found. "
"Install with: pip install torch==1.9.0 lpips==0.1.4. "
"LPIPS check will be skipped."
)
def compute_lpips(array1, array2):
"""
Compute LPIPS (Learned Perceptual Image Patch Similarity) between two arrays.
Lower values indicate more perceptually similar outputs.
Requires PyTorch and the LPIPS package.
"""
if torch is None or lpips_model is None:
return None
try:
# Cast arrays to numpy to ensure compatibility
array1_np = util.array.to_numpy(comp_util.cast_up(array1))
array2_np = util.array.to_numpy(comp_util.cast_up(array2))
# Check dimensions - LPIPS expects image data
shape1 = array1_np.shape
shape2 = array2_np.shape
# We need at least 3D arrays for LPIPS (typically B,C,H,W or H,W,C)
if len(shape1) < 3 or len(shape2) < 3:
G_LOGGER.warning(f"LPIPS requires at least 3D arrays, got shapes {shape1} and {shape2}. LPIPS check will be skipped.")
return None
# Log input shapes for debugging
G_LOGGER.verbose(f"Original tensor shapes: {shape1} and {shape2}")
# Get device information - use CPU for consistency
device = torch.device('cpu')
# Convert to PyTorch tensors with shape B,C,H,W
# LPIPS expects values in range [-1, 1] for both color and grayscale images
def prepare_for_lpips(arr):
# Determine input format and convert to B,C,H,W format
if len(arr.shape) == 3: # H,W,C or C,H,W
if arr.shape[2] <= 3: # H,W,C format
# Convert H,W,C to B,C,H,W (add batch dimension)
arr = arr.transpose(2, 0, 1)[None, ...]
else: # C,H,W format
# Add batch dimension
arr = arr[None, ...]
elif len(arr.shape) == 4: # B,C,H,W or B,H,W,C
if arr.shape[3] <= 3: # B,H,W,C format
arr = arr.transpose(0, 3, 1, 2)
# else: already in B,C,H,W format
# Convert to float and normalize to [-1, 1] range if needed
arr = arr.astype(np.float32)
if arr.max() > 1.0:
arr = arr / 255.0
if arr.max() <= 1.0 and arr.min() >= 0.0:
arr = arr * 2.0 - 1.0 # [0,1] -> [-1,1]
# Force 3-channel RGB format required by LPIPS
if arr.shape[1] == 1: # Grayscale (single channel)
# Repeat the channel 3 times to create RGB
arr = np.repeat(arr, 3, axis=1)
elif arr.shape[1] == 2: # Two channels
# Create a third channel (could duplicate channel 2 or create a new one)
third_channel = arr[:, 1:2] # Use second channel as the third
arr = np.concatenate([arr, third_channel], axis=1)
elif arr.shape[1] > 3: # More than 3 channels
arr = arr[:, :3] # Use only first 3 channels
# Convert to tensor
tensor = torch.from_numpy(arr)
return tensor.float().to(device)
# Convert both inputs to torch tensors in correct format
img1 = prepare_for_lpips(array1_np)
img2 = prepare_for_lpips(array2_np)
G_LOGGER.verbose(f"Prepared tensor shapes: {img1.shape} and {img2.shape}")
# Ensure tensors have the same size in all dimensions
if img1.shape != img2.shape:
G_LOGGER.warning(f"Tensor shapes don't match: {img1.shape} vs {img2.shape}. Adjusting...")
# For channels, ensure both have 3 channels
if img1.shape[1] != 3:
if img1.shape[1] == 1:
img1 = img1.repeat(1, 3, 1, 1)
elif img1.shape[1] == 2:
img1 = torch.cat([img1, img1[:, 1:2]], dim=1)
else: # > 3 channels
img1 = img1[:, :3]
if img2.shape[1] != 3:
if img2.shape[1] == 1:
img2 = img2.repeat(1, 3, 1, 1)
elif img2.shape[1] == 2:
img2 = torch.cat([img2, img2[:, 1:2]], dim=1)
else: # > 3 channels
img2 = img2[:, :3]
# For spatial dimensions, resize to match
if img1.shape[2:] != img2.shape[2:]:
# Use the larger of the two spatial dimensions
target_size = (max(img1.shape[2], img2.shape[2]), max(img1.shape[3], img2.shape[3]))
# Only import interpolate if needed
try:
from torch.nn.functional import interpolate
if img1.shape[2:] != target_size:
img1 = interpolate(img1, size=target_size, mode='bilinear', align_corners=False)
if img2.shape[2:] != target_size:
img2 = interpolate(img2, size=target_size, mode='bilinear', align_corners=False)
except ImportError:
G_LOGGER.warning("Failed to resize tensors: torch.nn.functional.interpolate not available")
if img1.shape[2:] != img2.shape[2:]:
G_LOGGER.warning("Cannot compute LPIPS with tensors of different spatial dimensions")
return None
G_LOGGER.verbose(f"Final tensor shapes: {img1.shape} and {img2.shape}")
# Make sure the model is in eval mode
lpips_model.eval()
# Compute LPIPS distance (using no_grad to avoid storing gradients)
with torch.no_grad():
try:
# Try the direct method
lpips_dist = lpips_model(img1, img2)
if isinstance(lpips_dist, torch.Tensor):
lpips_dist = lpips_dist.item()
except Exception as e:
G_LOGGER.warning(f"Standard LPIPS computation failed: {e}. Trying fallback method...")
try:
# Try an alternative approach
lpips_dist = lpips_model.forward(img1, img2)
if isinstance(lpips_dist, torch.Tensor):
lpips_dist = lpips_dist.mean().item()
except Exception as e2:
G_LOGGER.warning(f"Fallback LPIPS computation failed: {e2}. LPIPS check will be skipped.")
return None
return lpips_dist
except Exception as e:
G_LOGGER.warning(f"Error computing LPIPS: {e}. LPIPS check will be skipped.")
return None
def check_perceptual_metrics(
out0,
out0_name,
out1,
out1_name,
per_out_lpips_threshold,
runner0_name,
runner1_name,
):
"""
Checks whether two outputs match using perceptual metrics.
Args:
out0 (Union[np.array, torch.Tensor]): The first output.
out0_name (str): The name of the first output.
out1 (Union[np.array, torch.Tensor]): The second output.
out1_name (str): The name of the second output.
per_out_lpips_threshold (float): The maximum LPIPS score allowed for a match.
runner0_name (str): The name of the runner that generated the first output.
runner1_name (str): The name of the runner that generated the second output.
Returns:
PerceptualMetricsResult: Details on whether the outputs matched.
"""
# Log input information
G_LOGGER.super_verbose(
f"{runner0_name:35} | Output: {out0_name} (dtype={util.array.dtype(out0)}, shape={util.array.shape(out0)}):\n{util.indent_block(out0)}"
)
G_LOGGER.super_verbose(
f"{runner1_name:35} | Output: {out1_name} (dtype={util.array.dtype(out1)}, shape={util.array.shape(out1)}):\n{util.indent_block(out1)}"
)
# Compute LPIPS
lpips_value = compute_lpips(out0, out1)
# Check if outputs match based on the metrics
lpips_passed = True
if lpips_value is not None:
lpips_passed = bool(lpips_value <= per_out_lpips_threshold)
# Overall pass only depends on LPIPS for now
passed = lpips_passed
# Log information about the outputs
hist_bin_range = (
min(comp_util.compute_min(out0), comp_util.compute_min(out1)),
max(comp_util.compute_max(out0), comp_util.compute_max(out1)),
)
comp_util.log_output_stats(
out0, not passed, f"{runner0_name}: {out0_name}", hist_range=hist_bin_range
)
comp_util.log_output_stats(
out1, not passed, f"{runner1_name}: {out1_name}", hist_range=hist_bin_range
)
# Log perceptual metrics
G_LOGGER.info(f"Perceptual Metrics: {out0_name}")
with G_LOGGER.indent():
if lpips_value is not None:
G_LOGGER.info(f"LPIPS: {lpips_value:.5g} (max allowed: {per_out_lpips_threshold:.5g}) | {'PASSED' if lpips_passed else 'FAILED'}")
else:
G_LOGGER.warning("LPIPS computation was skipped or failed")
# Create a PerceptualMetricsResult object
result = PerceptualMetricsResult(
passed=passed,
lpips=lpips_value
)
# Log pass/fail status
if not passed:
if lpips_value is not None and not lpips_passed:
G_LOGGER.error(
f"FAILED | Output: '{out0_name}' | LPIPS ({lpips_value:.5g}) exceeds maximum threshold ({per_out_lpips_threshold:.5g})"
)
else:
metrics_passed = []
if lpips_value is not None:
metrics_passed.append("LPIPS")
if metrics_passed:
G_LOGGER.finish(
f"PASSED | Output: '{out0_name}' | All perceptual metrics passed: {', '.join(metrics_passed)}"
)
else:
G_LOGGER.warning(
f"PASSED | Output: '{out0_name}' | No perceptual metrics were successfully computed"
)
return result
def compare_output(iter_result0, iter_result1):
"""
Compare the outputs of two runners from a single iteration using perceptual metrics.
This function will always iterate over the output names of the first IterationResult,
and attempt to find corresponding output names in the second.
If no corresponding output name is found, the output is skipped.
If all output names are skipped, then this function raises an error.
Args:
iter_result0 (IterationResult): The result of the first runner.
iter_result1 (IterationResult): The result of the second runner.
Returns:
OrderedDict[str, PerceptualMetricsResult]:
The name of the outputs compared, derived from the first IterationResult,
and whether they matched. If an output name is not found, it is omitted from this dictionary.
Raises:
PolygraphyException: If all output names are skipped, and thus no outputs are compared.
"""
def check_dict(dct, dict_name):
if isinstance(dct, dict):
util.check_sequence_contains(
dct.keys(),
set(iter_result0.keys()) | set(iter_result1.keys()) | {""},
name=dict_name,
log_func=G_LOGGER.warning,
check_missing=False,
)
check_dict(lpips_threshold, "the lpips_threshold dictionary")
if not check_shapes:
G_LOGGER.info(
"Strict shape checking disabled. Will attempt to match output shapes before comparisons"
)
def match(out0_name, output0, out1_name, output1):
per_out_lpips_threshold = util.value_or_from_dict(lpips_threshold, out0_name, default_lpips_threshold)
# Log threshold information
G_LOGGER.info(f"Perceptual Tolerance: [LPIPS max={per_out_lpips_threshold:.5g}]")
G_LOGGER.extra_verbose(
f"Note: Comparing {iter_result0.runner_name} vs. {iter_result1.runner_name}"
)
if check_shapes and util.array.shape(output0) != util.array.shape(
output1
):
G_LOGGER.error(
f"FAILED | Output: `{out0_name}` | Will not compare outputs of different shapes.\n"
f"Note: Output shapes are {util.array.shape(output0)} and {util.array.shape(output1)}."
)
G_LOGGER.error(
"Note: Use --no-shape-check or set check_shapes=False to "
"attempt to compare values anyway.",
mode=LogMode.ONCE,
)
return False
output1 = util.try_match_shape(output1, util.array.shape(output0))
output0 = util.array.view(
output0,
DataType.from_dtype(util.array.dtype(output0)),
util.array.shape(output1),
)
outputs_matched = check_perceptual_metrics(
output0,
out0_name,
output1,
out1_name,
per_out_lpips_threshold=per_out_lpips_threshold,
runner0_name=iter_result0.runner_name,
runner1_name=iter_result1.runner_name,
)
return outputs_matched
nonlocal find_output_func
find_output_func = util.default(
find_output_func,
functools.partial(
default_find_output_func, base_iter_result=iter_result0
),
)
return run_comparison(
match, fail_fast, iter_result0, iter_result1, find_output_func
)
return compare_output