New candidate spaces (bidirectional; forward and backward)

[dilpath]

• bidirectional candidate space
  • moves as far in one direction before no further models are found, then moves switches, until switching twice in a row yields no good models
• forward and backward candidate space
  • considers all models that are a valid forward or backward move
  • currently might be that, e.g., the closest "forward models" are +1 estimated parameters, and the closest "backward models" are -3 estimated parameters: these models are part of the same iteration
• candidate spaces can now implement a wrapper for the subspaces search
  • implemented for the direction switches in the new candidate spaces
• added ModelSpace.from_df
• added test for bidirectional candidate space
  • for this test: added a parameter to the example problem in doc/examples/model_selection

[codecov-commenter]

Codecov Report

:exclamation: No coverage uploaded for pull request base (develop@9691901). Click here to learn what that means. The diff coverage is n/a.

@@            Coverage Diff             @@
##             develop      #17   +/-   ##
==========================================
  Coverage           ?   78.61%           
==========================================
  Files              ?       14           
  Lines              ?     1141           
  Branches           ?        0           
==========================================
  Hits               ?      897           
  Misses             ?      244           
  Partials           ?        0           

---

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 9691901...9bf42c0. Read the comment docs.