biolab / orange3-timeseries

🍊 :chart_with_upwards_trend: Orange add-on for analyzing, visualizing, manipulating, and forecasting time series data.

Other

62 stars 41 forks source link

Updates for newer dependencies #195

Closed janezd closed 2 years ago

janezd commented 2 years ago

Issue

There have been multiple changes in statsmodels, as well as in Orange, for instance table locking.

Also fixes #196.

Description of changes

I unlocked tables where necessary: TimeSeries was unlockable because it was constructed as a view into an intermediate table
I increase required Orange version to 3.31 and removed pins (otherwise installation failed due to conflicts)
argument 'unbiased' in call of autocorrelation was removed because it is no longer available; there have been some arguments against "unbiasing" models in this way.
freq has been renamed to period; meaning is the same, I've found the statsmodels's commit in which they renamed it
Model Evaluation contained an invalid fix (#37): if a classifier errored, some function returned a list with 6 "err" strings instead of 7; numpy couldn't create a proper 2d array and created a 1d array of lists. This was "fixed" so that a widget checked ndim and if it was 1, printed a warning "something went wrong". 🤯 I added a missing "err" (which was, as I see, all that would need to be done in #37!, but then still kept a rephrased warning about errored model(s).

Includes

[X] Code changes
[X] Tests

janezd commented 2 years ago

Most constructors for Timeseries are patched through Timeseries.from_data_table, which construct a view. This is mostly unnecessary in happens only for convience. However, the result is that those table can't be unlocked.

@markotoplak, what you said on Friday can really bite. In this particular case, though, I think it should only affect tests.

In general, we should strive to write widgets so that they never modify a table once it's created. Patching should be limited to tests.

codecov-commenter commented 2 years ago

Codecov Report

Merging #195 (9b70f11) into master (3ce4ca1) will increase coverage by 0.76%. The diff coverage is 91.66%.

@@            Coverage Diff             @@
##           master     #195      +/-   ##
==========================================
+ Coverage   72.43%   73.20%   +0.76%     
==========================================
  Files           7        7              
  Lines         809      821      +12     
  Branches      133      161      +28     
==========================================
+ Hits          586      601      +15     
+ Misses        171      168       -3     
  Partials       52       52

Impacted Files	Coverage Δ
orangecontrib/timeseries/timeseries.py	`85.95% <87.50%> (+0.83%)`	:arrow_up:
orangecontrib/timeseries/functions.py	`77.09% <100.00%> (ø)`
orangecontrib/timeseries/models.py	`69.60% <100.00%> (+1.78%)`	:arrow_up:

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 14a557f...9b70f11. Read the comment docs.

janezd commented 2 years ago

@markotoplak, can you review this? In particular because of https://github.com/biolab/orange3-timeseries/pull/195/commits/7ef044f31ab11d2cf80aa15f0bbcc5a80308c44e.

Changes are explained in bullets in description.

Merging this (if correct) is necessary to continue any work on this add-on.