Return expected number of values in compute_prebins for ContinuousOptimalBinning

[YC-1412]

Fixes #357.

Please see the context in #357. This PR make _compute_prebins in ContinuousOptimalBinning always return 8 values. Missing values are filled with empty numpy array.

I used the test code in #357 to test the fix.

Before

After

[YC-1412]

Hi @guillermo-navas-palencia, I noticed the pytest workflow failed on test_binning_piecewise.py. It seems to come from the osqp version (one dependency ofcvxpy). When I test it locally, it fails on the main branch code as well.

When I use osqp==0.6.7.post2, the test passes, but it fails when osqp>1.0. It produces a different optimizer.coef_ from theRobustPWRegression optimizer while the input x and y are unchanged.

• optimizer.coef_:
  

  optbinning/optbinning/binning/piecewise/base.py

  Lines 459 to 476 in 0720b8d

  	optimizer = RobustPWRegression(
  	objective=self.objective,
  	degree=self.degree,
  	continuous=self.continuous,
  	continuous_deriv=self.continuous_deriv,
  	monotonic_trend=monotonic,
  	solver=self.solver,
  	h_epsilon=self.h_epsilon,
  	quantile=self.quantile,
  	regularization=self.regularization,
  	reg_l1=self.reg_l1,
  	reg_l2=self.reg_l2,
  	extrapolation="continue",
  	verbose=self.verbose)
  	optimizer.fit(x_subsamples, pred_subsamples, splits, lb=lb, ub=ub)
  	self._c = optimizer.coef_

  
  

• test result:
  

Unfortunately, I'm not sure about the root cause and version difference in osqp. The results look close to me so it might be some rounding update. It passes the test if setting the tolerance to $10^{-5}$ or updating the expected value to the result of the newer version (5.84465065). Please let me know if you think one of them sounds a reasonable fix.

[guillermo-navas-palencia]

Hi @YC-1412. I think setting a tolerance of 1e-4 is enough. I have had similar problems with previous cvxpy updates/changes.

[YC-1412]

Hi @guillermo-navas-palencia, thanks for the confirmation. I took the liberty to update the tolerance in this PR. If it's not appropriate, I can revert it. Thanks!

[guillermo-navas-palencia]

The tolerance change for 1 test was missing. I think after this change, all tests will pass :)

[YC-1412]

Updated! Thank you!