Fix format of values in confusion matrix plot. #16159

Rick-Mackenbach · 2020-01-20T09:31:26Z

Reference Issues/PRs

In reference to this, #16127

What does this implement/fix? Explain your changes.

The formatting of values in the plot of the confusion matrix were set to '.2g' if no other format was specified. As of now, if no format is provided by the user, the format will be 'd' if the value is <1e7, and '.2g' otherwise.

fix will now show values up to 1e7 in integer values, and afterwards in .2g"

Rick-Mackenbach · 2020-01-20T15:15:44Z

@jnothman Hey Joel, I made the fix name a bit clearer now.

Could you help me understand why the codecov/patch and codecov/project tests are not passing?

to check for array_equal."

counted in the if statement."

Rick-Mackenbach · 2020-01-21T10:32:23Z

Well that certainly was a journey, but it seems to be working now.

@NicolasHug What do you think?

NicolasHug

Thanks @Rick-Mackenbach , made a few comments

Just FYI, you don't need to close and open new PR, you can simply push changes to your already-existing PR. It makes it easier for us to keep track of it ;)

NicolasHug · 2020-01-21T12:44:39Z

sklearn/metrics/_plot/confusion_matrix.py

-                                           ha="center", va="center",
-                                           color=color)
+
+            if (values_format is None):


No parenthesis

NicolasHug · 2020-01-21T12:44:39Z

sklearn/metrics/_plot/confusion_matrix.py

+                    color = cmap_max if cm[i, j] < thresh else cmap_min
+
+                    self.text_[i, j] = ax.text(
+                        j, i, format(cm[i, j], values_format),
+                        ha="center", va="center",
+                        color=color)


These lines are duplicated.

Let's put the if value_format is None condition inside of the for loop to avoid duplication.

NicolasHug · 2020-01-21T12:44:39Z

sklearn/metrics/_plot/confusion_matrix.py

+        if (values_format is None and cm.dtype.kind == "f"):
+            values_format = '.2g'


merge this condition with the rest below? It's better to have all related logic at the same place

NicolasHug · 2020-01-21T12:44:39Z

sklearn/metrics/_plot/tests/test_plot_confusion_matrix.py

-    text_text = np.array([
-        t.get_text() for t in disp.text_.ravel()])
-    assert_array_equal(expected_text, text_text)
+    if values_format is None:


Having a big if/else statement like this kinda defeats the purpose of parametrizing the test.

Maybe keep this test unchanged and create a new one test_confusion_matrix_text_format_default which would only check the default behavior.

inside the ConfusionMatrixDisplay."

…ikit-learn into confusion_fix

jnothman · 2020-02-12T09:42:25Z

I don't think that merge worked well... The diff now includes all changes.

You should have been able to do something like:

git checkout master
git pull https://github.com/scikit-learn/scikit-learn master
git checkout my_branch
git merge master
git push

not sure that'll work now. You might want to reset back to your last good commit using git reset --hard. But it seems your commits and the commits in master are now interleaved...

It might be better to lose your history by doing something like:

git checkout confusion_fix
git reset --hard master
git checkout master sklearn/metrics/_plot/confusion_matrix.py sklearn/metrics/_plot/tests/test_plot_confusion_matrix.py  # are these the only files you intended to modify in this PR??
git status
git commit

…ikit-learn into confusion_fix

Rick-Mackenbach · 2020-02-12T10:56:16Z

Thanks @jnothman for the fix!

jnothman · 2020-02-12T12:02:52Z

I hope it kept all your changes!

Rick-Mackenbach · 2020-02-12T12:41:35Z

I hope it kept all your changes!

Seems like it did from what I can see :)

thomasjpfan

This looks almost done!

thomasjpfan · 2020-02-20T20:26:49Z

sklearn/metrics/_plot/confusion_matrix.py

-                                           color=color)
+
+                fmt = values_format
+                if values_format is None:


I think trying .2g and then trying d if applicable, is slightly clearer

This suggestion is to address #16159 (comment)

if values_format is None: text_cm = format(cm[i, j], '.2g') if cm.dtype.kind != 'f': text_d = format(cm[i, j], 'd') if len(text_d) < len(text_cm): text_cm = text_d else: text_cm = format(cm[i, j], values_format)

The test would need to updated to reflect this behavior.

Rick-Mackenbach · 2020-02-21T14:26:43Z

@thomasjpfan Not sure if this is what you meant?

thomasjpfan

Thank you for your continued work and patience on this PR.

thomasjpfan · 2020-02-21T19:20:31Z

sklearn/metrics/_plot/confusion_matrix.py

+                                                        format(cm[i, j], 'd')):
+                            fmt = '.2g'
+
+                self.text_[i, j] = ax.text(


Sorry I was not more clear. The current PR may call format three times. I am trying to push for calling it at most twice:

if values_format is None: text_cm = format(cm[i, j], '.2g') if cm.dtype.kind != 'f': text_d = format(cm[i, j], 'd') if len(text_d) < len(text_cm): text_cm = text_d else: text_cm = format(cm[i, j], values_format) self.text_[i, j] = ax.text(j, i, text_cm, ...)

Rick-Mackenbach · 2020-02-24T14:04:45Z

@thomasjpfan My pleasure, thanks for you continuous feedback on this :).

I wasn't looking correctly regarding the correction, I see now that the formatting is done in the if-statement, and you just pass this formatted text to self.text_

Should be correct now 👍

thomasjpfan

LGTM

thomasjpfan · 2020-02-24T17:50:48Z

Please add an entry to the change log at doc/whats_new/v0.23.rst with a API tag. Like the other entries there, please reference this pull request with :pr: and credit yourself (and other contributors if applicable) with :user:.

Rick-Mackenbach · 2020-02-27T16:06:19Z

@thomasjpfan Not sure why the tests have failed here, could you help me out a bit? :)

thomasjpfan · 2020-02-29T20:20:09Z

I pushed a change to fix up the PR. It should be ready now.

jnothman · 2020-03-01T03:07:41Z

Thanks @Rick-Mackenbach!

Rick Mackenbach added 19 commits January 17, 2020 13:56

"Changes to the values_format for plotting the confusion_matrix. This

f3ee74f

fix will now show values up to 1e7 in integer values, and afterwards in .2g"

"Fixes for confusion_matrix values_format."

5fb226b

"<= value for 0"

cb0083e

"Changed back to == 0"

551efa8

"Import math log10 for linting error"

e253373

"Linting and hard-coded value checks now working"

02995c5

"Use log(10) instead of 1e7 (python version error?)"

e8019fb

"If cmd.type == f test included"

8d6f8a3

"Added type test in the correct place"

3b28b17

"Different IF statement, see if it fixes azure pipeline errors"

6208f85

"Changed order and use np.log10 ofcourse"

c32cc01

"Removed brackets"

18137c1

"Added testing in test_plot_confusion_matrix (values_format)"

409c251

"Wrong value format"

3e7168e

"Changed the test_plot_confusion_matrix."

23b3a20

"Add test plot confusion matrix"

d31518a

"Added extra test cases for test_plot_confusion_matrix"

5e9b925

"Fixed testing values"

46fef3d

"Added argument to the test function, duh"

0c9cb36

Rick Mackenbach added 5 commits January 20, 2020 17:38

"New tests to also test if for values > 1e7"

d21a7ff

"New test version, made sure the values are now loaded into the model

93cdb7a

to check for array_equal."

"Improved test file"

af47489

"New changes"

c404c77

"Fixed the ConfusionMatrixDisplay class, where values where not

1f766ca

counted in the if statement."

NicolasHug reviewed Jan 21, 2020

View changes

Rick Mackenbach added 3 commits January 22, 2020 09:27

"Fixing of loop and test"

690736f

"Changed if statement in the for loop and new test function"

89e6d1a

"Changes to test function and added the if condition in the for-loop

ff0cd1a

inside the ConfusionMatrixDisplay."

Merge branch 'confusion_fix' of https://github.com/Rick-Mackenbach/sc…

b80d38c

…ikit-learn into confusion_fix

Rick Mackenbach added 2 commits February 12, 2020 11:28

"Fix of wrong merge"

74f7244

Merge branch 'confusion_fix' of https://github.com/Rick-Mackenbach/sc…

a71357d

…ikit-learn into confusion_fix

Rick-Mackenbach requested a review from thomasjpfan February 17, 2020 07:46

"Small change in test file"

6c37b11

thomasjpfan reviewed Feb 20, 2020

View changes

"Updated the logic for interpretability. Also small changes to test"

19319e6

thomasjpfan reviewed Feb 21, 2020

View changes

"Do the formatting of values in the if-statement, so shorter code :)"

de4e76e

"Changed test file to test for 'd' values"

666246d

thomasjpfan approved these changes Feb 24, 2020

View changes

Rick Mackenbach added 2 commits February 25, 2020 17:30

"Added changes to the v0.23 what's news"

7895260

"Modified doc/whats_new/v0.23.rst"

e920a7b

Merge remote-tracking branch 'upstream/master' into pr/16159

3365587

jnothman approved these changes Mar 1, 2020

View changes

jnothman merged commit 94f877b into scikit-learn:master Mar 1, 2020

Rick-Mackenbach deleted the confusion_fix branch March 2, 2020 10:17

panpiort8 pushed a commit to panpiort8/scikit-learn that referenced this pull request Mar 3, 2020

Fix format of values in confusion matrix plot. (scikit-learn#16159)

c62f829

cmarmo mentioned this pull request Mar 5, 2020

formatting of values in plot_confusion_matrix #16127

Closed

gio8tisu pushed a commit to gio8tisu/scikit-learn that referenced this pull request May 15, 2020

Fix format of values in confusion matrix plot. (scikit-learn#16159)

848f69a

Fix format of values in confusion matrix plot. #16159

Fix format of values in confusion matrix plot. #16159

Rick-Mackenbach commented Jan 20, 2020

Rick-Mackenbach commented Jan 20, 2020

Rick-Mackenbach commented Jan 21, 2020 •

edited

NicolasHug left a comment

NicolasHug Jan 21, 2020

NicolasHug Jan 21, 2020

NicolasHug Jan 21, 2020

NicolasHug Jan 21, 2020

jnothman commented Feb 12, 2020

Rick-Mackenbach commented Feb 12, 2020

jnothman commented Feb 12, 2020 via email

Rick-Mackenbach commented Feb 12, 2020

thomasjpfan left a comment

thomasjpfan Feb 20, 2020

Rick-Mackenbach commented Feb 21, 2020

thomasjpfan left a comment

thomasjpfan Feb 21, 2020

Rick-Mackenbach commented Feb 24, 2020

thomasjpfan left a comment

thomasjpfan commented Feb 24, 2020

Rick-Mackenbach commented Feb 27, 2020

thomasjpfan commented Feb 29, 2020

jnothman commented Mar 1, 2020

		if (values_format is None and cm.dtype.kind == "f"):
		values_format = '.2g'

Fix format of values in confusion matrix plot. #16159

Fix format of values in confusion matrix plot. #16159

Conversation

Rick-Mackenbach commented Jan 20, 2020

Reference Issues/PRs

What does this implement/fix? Explain your changes.

Rick-Mackenbach commented Jan 20, 2020

Rick-Mackenbach commented Jan 21, 2020 • edited

NicolasHug left a comment

Choose a reason for hiding this comment

NicolasHug Jan 21, 2020

Choose a reason for hiding this comment

NicolasHug Jan 21, 2020

Choose a reason for hiding this comment

NicolasHug Jan 21, 2020

Choose a reason for hiding this comment

NicolasHug Jan 21, 2020

Choose a reason for hiding this comment

jnothman commented Feb 12, 2020

Rick-Mackenbach commented Feb 12, 2020

jnothman commented Feb 12, 2020 via email

Rick-Mackenbach commented Feb 12, 2020

thomasjpfan left a comment

Choose a reason for hiding this comment

thomasjpfan Feb 20, 2020

Choose a reason for hiding this comment

Rick-Mackenbach commented Feb 21, 2020

thomasjpfan left a comment

Choose a reason for hiding this comment

thomasjpfan Feb 21, 2020

Choose a reason for hiding this comment

Rick-Mackenbach commented Feb 24, 2020

thomasjpfan left a comment

Choose a reason for hiding this comment

thomasjpfan commented Feb 24, 2020

Rick-Mackenbach commented Feb 27, 2020

thomasjpfan commented Feb 29, 2020

jnothman commented Mar 1, 2020

Rick-Mackenbach commented Jan 21, 2020 •

edited