Skip to content

Fix PR curve plugin to display correct thresholds #5191

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
nfelt merged 2 commits into from
Aug 10, 2021
Merged

Fix PR curve plugin to display correct thresholds #5191

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4bde868
Fix PR curve plugin to display correct thresholds
dcower Aug 4, 2021
e14c54b
Fix linter error in _make_pr_entry
dcower Aug 7, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion tensorboard/plugins/pr_curve/pr_curve_demo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -267,7 +267,7 @@ def run_all(logdir, steps, thresholds, verbose=False):

def main(unused_argv):
print("Saving output to %s." % FLAGS.logdir)
run_all(FLAGS.logdir, FLAGS.steps, 50, verbose=True)
run_all(FLAGS.logdir, FLAGS.steps, 51, verbose=True)
print("Done. Output saved to %s." % FLAGS.logdir)


Expand Down
34 changes: 16 additions & 18 deletions tensorboard/plugins/pr_curve/pr_curves_plugin.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,23 +205,27 @@ def _make_pr_entry(self, step, wall_time, data_array):
Returns:
A PR curve entry.
"""
# Trim entries for which TP + FP = 0 (precision is undefined) at the tail of
# the data.
true_positives = [
int(v) for v in data_array[metadata.TRUE_POSITIVES_INDEX]
]
false_positives = [
int(v) for v in data_array[metadata.FALSE_POSITIVES_INDEX]
]
tp_index = metadata.TRUE_POSITIVES_INDEX
fp_index = metadata.FALSE_POSITIVES_INDEX
tn_index = metadata.TRUE_NEGATIVES_INDEX
fn_index = metadata.FALSE_NEGATIVES_INDEX

# Trim entries for which TP + FP = 0 (precision is undefined) at the tail of
# the data.
positives = data_array[[tp_index, fp_index], :].astype(int).sum(axis=0)
# Searching from the end, find the farthest index where TP + FP = 0.
end_index_inclusive = len(positives) - 1
while end_index_inclusive > 0 and positives[end_index_inclusive] == 0:
end_index_inclusive -= 1
end_index = end_index_inclusive + 1
# Generate thresholds in [0, 1].
num_thresholds = data_array.shape[1]
thresholds = (np.arange(1, end_index + 1) / num_thresholds).tolist()
thresholds = np.linspace(0.0, 1.0, num_thresholds)

true_positives = [int(v) for v in data_array[tp_index]]
false_positives = [int(v) for v in data_array[fp_index]]
true_negatives = [int(v) for v in data_array[tn_index]]
false_negatives = [int(v) for v in data_array[fn_index]]

return {
"wall_time": wall_time,
Expand All @@ -232,13 +236,7 @@ def _make_pr_entry(self, step, wall_time, data_array):
"recall": data_array[metadata.RECALL_INDEX, :end_index].tolist(),
"true_positives": true_positives[:end_index],
"false_positives": false_positives[:end_index],
"true_negatives": [
int(v)
for v in data_array[metadata.TRUE_NEGATIVES_INDEX][:end_index]
],
"false_negatives": [
int(v)
for v in data_array[metadata.FALSE_NEGATIVES_INDEX][:end_index]
],
"thresholds": thresholds,
"true_negatives": true_negatives[:end_index],
"false_negatives": false_negatives[:end_index],
"thresholds": thresholds[:end_index].tolist(),
}
12 changes: 6 additions & 6 deletions tensorboard/plugins/pr_curve/pr_curves_plugin_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,7 +214,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[300, 201, 38, 2],
expected_true_negatives=[0, 99, 262, 298],
expected_false_negatives=[0, 24, 105, 144],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[0],
)
self.validatePrCurveEntry(
Expand All @@ -225,7 +225,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[300, 204, 39, 6],
expected_true_negatives=[0, 96, 261, 294],
expected_false_negatives=[0, 22, 105, 146],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[1],
)
self.validatePrCurveEntry(
Expand All @@ -236,7 +236,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[300, 185, 38, 2],
expected_true_negatives=[0, 115, 262, 298],
expected_false_negatives=[0, 30, 111, 146],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[2],
)

Expand All @@ -252,7 +252,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[150, 105, 18, 0],
expected_true_negatives=[0, 45, 132, 150],
expected_false_negatives=[0, 11, 54, 70],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[0],
)
self.validatePrCurveEntry(
Expand All @@ -263,7 +263,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[150, 99, 21, 3],
expected_true_negatives=[0, 51, 129, 147],
expected_false_negatives=[0, 13, 54, 74],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[1],
)
self.validatePrCurveEntry(
Expand All @@ -274,7 +274,7 @@ def testPrCurvesDataCorrect(self):
expected_false_positives=[150, 92, 20, 1],
expected_true_negatives=[0, 58, 130, 149],
expected_false_negatives=[0, 14, 59, 73],
expected_thresholds=[0.2, 0.4, 0.6, 0.8],
expected_thresholds=[0.0, 0.25, 0.5, 0.75],
pr_curve_entry=entries[2],
)

Expand Down
63 changes: 63 additions & 0 deletions tensorboard/plugins/pr_curve/summary_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -285,6 +285,69 @@ def test_counts_below_1(self):
values = tensor_util.make_ndarray(pb.value[0].tensor)
self.verify_float_arrays_are_equal(expected, values)

def test_predictions_at_thresholds(self):
num_thresholds = 5
# For 5 thresholds, we expect the thresholds used to be:
# [0.0, 0.25, 0.5, 0.75, 1.0]
pb = self.compute_and_check_summary_pb(
name="foo",
labels=np.array([True] * num_thresholds),
predictions=np.float32([0.0, 0.25, 0.5, 0.75, 1.0]),
num_thresholds=num_thresholds,
)
expected = [
[5.0, 4.0, 3.0, 2.0, 1.0], # TP
[0.0, 0.0, 0.0, 0.0, 0.0], # FP
[0.0, 0.0, 0.0, 0.0, 0.0], # TN
[0.0, 1.0, 2.0, 3.0, 4.0], # FN
[1.0, 1.0, 1.0, 1.0, 1.0], # Precision
[1.0, 0.8, 0.6, 0.4, 0.2], # Recall
]
values = tensor_util.make_ndarray(pb.value[0].tensor)
self.verify_float_arrays_are_equal(expected, values)

def test_predictions_above_thresholds(self):
num_thresholds = 5
# For 5 thresholds, we expect the thresholds used to be:
# [0.0, 0.25, 0.5, 0.75, 1.0]
pb = self.compute_and_check_summary_pb(
name="foo",
labels=np.array([True] * num_thresholds),
predictions=np.float32([0.01, 0.26, 0.51, 0.76, 1.0]),
num_thresholds=num_thresholds,
)
expected = [
[5.0, 4.0, 3.0, 2.0, 1.0], # TP
[0.0, 0.0, 0.0, 0.0, 0.0], # FP
[0.0, 0.0, 0.0, 0.0, 0.0], # TN
[0.0, 1.0, 2.0, 3.0, 4.0], # FN
[1.0, 1.0, 1.0, 1.0, 1.0], # Precision
[1.0, 0.8, 0.6, 0.4, 0.2], # Recall
]
values = tensor_util.make_ndarray(pb.value[0].tensor)
self.verify_float_arrays_are_equal(expected, values)

def test_predictions_below_thresholds(self):
num_thresholds = 5
# For 5 thresholds, we expect the thresholds used to be:
# [0.0, 0.25, 0.5, 0.75, 1.0]
pb = self.compute_and_check_summary_pb(
name="foo",
labels=np.array([True] * num_thresholds),
predictions=np.float32([0.0, 0.24, 0.49, 0.74, 0.99]),
num_thresholds=num_thresholds,
)
expected = [
[5.0, 3.0, 2.0, 1.0, 0.0], # TP
[0.0, 0.0, 0.0, 0.0, 0.0], # FP
[0.0, 0.0, 0.0, 0.0, 0.0], # TN
[0.0, 2.0, 3.0, 4.0, 5.0], # FN
[1.0, 1.0, 1.0, 1.0, 0.0], # Precision
[1.0, 0.6, 0.4, 0.2, 0.0], # Recall
]
values = tensor_util.make_ndarray(pb.value[0].tensor)
self.verify_float_arrays_are_equal(expected, values)

def test_raw_data(self):
# We pass these raw counts and precision/recall values.
name = "foo"
Expand Down