Fix orig edivisive flow

[vishnuchalla]

Description

Fix orig edivsive flow. I am running into below error if more than 10 datapoints are being used for analysis

vchalla@vchalla-thinkpadp1gen2:~/myforks/otava$ otava --config-file /tmp/otava-10.yaml analyze tiny10 --last 15 -P 0.001 --window 15 -M 0.0 --orig-edivisive=true
INFO: Computing change points for test tiny10...
Traceback (most recent call last):
  File "/home/vchalla/.local/bin/otava", line 7, in <module>
    sys.exit(main())
             ^^^^^^
  File "/home/vchalla/myforks/otava/otava/main.py", line 651, in main
    script_main()
  File "/home/vchalla/myforks/otava/otava/main.py", line 578, in script_main
    analyzed_series = otava.analyze(
                      ^^^^^^^^^^^^^^
  File "/home/vchalla/myforks/otava/otava/main.py", line 125, in analyze
    produced_report = report.produce_report(test.name, report_type)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/vchalla/myforks/otava/otava/report.py", line 52, in produce_report
    return self.__format_log_annotated(test_name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/vchalla/myforks/otava/otava/report.py", line 87, in __format_log_annotated
    change = [c for c in cp.changes if c.metric == col_name]
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/vchalla/myforks/otava/otava/report.py", line 87, in <listcomp>
    change = [c for c in cp.changes if c.metric == col_name]
                                       ^^^^^^^^
AttributeError: 'ChangePoint' object has no attribute 'metric'

Testing

Tested and verified in local

vchalla@vchalla-thinkpadp1gen2:~/myforks/otava$ otava --config-file /tmp/otava-15.yaml analyze tiny15 -P 0.001 --window 50 -M 0.0 --orig-edivisive
INFO: Computing change points for test tiny15...
time                       commit      metric1
-------------------------  --------  ---------
2026-01-01 00:00:00 +0000  c1              100
2026-01-02 00:00:00 +0000  c2              101
2026-01-03 00:00:00 +0000  c3               99
2026-01-04 00:00:00 +0000  c4              100
2026-01-05 00:00:00 +0000  c5              102
2026-01-06 00:00:00 +0000  c6              101
2026-01-07 00:00:00 +0000  c7              100
2026-01-08 00:00:00 +0000  c8               99
2026-01-09 00:00:00 +0000  c9              100
                                     ·········  
                                        +30.3%  
                                     ·········  
2026-01-10 00:00:00 +0000  c10             130
2026-01-11 00:00:00 +0000  c11             131
2026-01-12 00:00:00 +0000  c12             130
2026-01-13 00:00:00 +0000  c13             132
2026-01-14 00:00:00 +0000  c14             130

[vishnuchalla]

cc: @henrikingo @Gerrrr

[henrikingo]

Unfortunately, rather than fixing the orig_edivisive flow, in master, we should actually remove it.

By orig_edivisive we mean the original, close to textbook implementation at MongoDB in 2017. Since we no longer want to depend on the external dependency, we cannot offer this option anymore.

Note that in the 0.7 branch you can still use --orig-edivisive, but as far as I can tell, this bug isn't happening in that branch.

[Gerrrr]

Unfortunately, rather than fixing the orig_edivisive flow, in master, we should actually remove it.

+1.

Note that in the 0.7 branch you can still use --orig-edivisive, but as far as I can tell, this bug isn't happening in that branch.

@vishnuchalla if you find issues with --orig-edivisive in 0.7.0, we can fix it there and do a bugfix release.

[henrikingo]

Closing. If we misunderstood something, please just reply here and reopen.

[henrikingo]

Oh dear... My sincerest apologies. I was looking into something unrelated yesterday and realized that there is a completely valid --orig-edivisive flow also in the new algorithm in main branch / 0.8.0 and higher. I had forgotten that @Sowiks did re-implement a version of the algorithm that is entered at

otava/otava/analysis.py

Line 280 in 1ceb153

def compute_change_points_orig(series: Sequence[SupportsFloat], max_pvalue: float = 0.001, seed: Optional[int] = None) -> Tuple[PermCPList, Optional[PermCPList]]:

...and your patch is against that.

I will re-open this and look at your patch later.

[henrikingo]

Ok yes,I caught up with this.

Working indepently on a different patch, I end up needing exactly this patch for --orig-edivisive

+1

[henrikingo]

I think your patch looks good except for this one linting failure. If you move the TTestSignificanceTester one line up, I'll hit the button to rerun tests for you.

In the mean time, do you already have a signed contributor agreement on file with the ASF? If not:

https://www.apache.org/licenses/contributor-agreements.html
Short version is, download the ICLA pdf, sign it either with a pen or GnuPG, email it to secretary@apache.org

[vishnuchalla]

I think your patch looks good except for this one linting failure. If you move the TTestSignificanceTester one line up, I'll hit the button to rerun tests for you.

In the mean time, do you already have a signed contributor agreement on file with the ASF? If not:

https://www.apache.org/licenses/contributor-agreements.html
Short version is, download the ICLA pdf, sign it either with a pen or GnuPG, email it to secretary@apache.org

@henrikingo Singed and emailed the ICLA. Also fixed the linting issue. PTAL once you get sometime. Thank you.

[henrikingo]

Whether testing the full output of otava analyze -h is meaningful is another question... But anyway, you need

--- a/tests/cli_help_test.py
+++ b/tests/cli_help_test.py

@@ -162,7 +162,7 @@ usage: otava analyze [-h] [--config-file CONFIG_FILE] [--graphite-url GRAPHITE_U
                      [--output {{log,json,regressions_only}}] [--branch [STRING]] [--metrics LIST]
 {usage_filter_lines}
                      [--last COUNT] [-P, --p-value PVALUE] [-M MAGNITUDE] [--window WINDOW]
-                     [--orig-edivisive ORIG_EDIVISIVE]
+                     [--orig-edivisive]
                      tests [tests ...]
 
-  --orig-edivisive ORIG_EDIVISIVE
-                        use the original edivisive algorithm with no windowing and weak change
+  --orig-edivisive      use the original edivisive algorithm with no windowing and weak change
                         points analysis improvements

[vishnuchalla]

Whether testing the full output of otava analyze -h is meaningful is another question... But anyway, you need

--- a/tests/cli_help_test.py
+++ b/tests/cli_help_test.py

@@ -162,7 +162,7 @@ usage: otava analyze [-h] [--config-file CONFIG_FILE] [--graphite-url GRAPHITE_U
                      [--output {{log,json,regressions_only}}] [--branch [STRING]] [--metrics LIST]
 {usage_filter_lines}
                      [--last COUNT] [-P, --p-value PVALUE] [-M MAGNITUDE] [--window WINDOW]
-                     [--orig-edivisive ORIG_EDIVISIVE]
+                     [--orig-edivisive]
                      tests [tests ...]
 
-  --orig-edivisive ORIG_EDIVISIVE
-                        use the original edivisive algorithm with no windowing and weak change
+  --orig-edivisive      use the original edivisive algorithm with no windowing and weak change
                         points analysis improvements

I think the tests just need a re-run? Updated with this patch.

[henrikingo]

The merge re-introduced 413d774

[vishnuchalla]

The merge re-introduced 413d774

Oops, fixed.

[henrikingo]

Thanks. And nice to have you here @vishnuchalla !