ENH Add streaming functionality to Synergistic Forest

[nhahn7]

Reference issue

#34 #501

Type of change

Feature Request

What does this implement/fix?

Add streaming functionality for forests with update_task function.

Additional information

• For streaming Synergistic Forests, using update_task requires partial fitting from scikit-learn-stream fork

• Notebook demonstrating SynF vs DF (streaming and batch)

• Extended from Add Streaming Function #526 Added streaming capability for ODIN #528

[PSSF23]

Remember to remove the extra print statements~

[PSSF23]

black formatting is not complete.

[codecov]

Codecov Report

Merging #537 (87fab73) into staging (af84f50) will decrease coverage by 13.40%.
The diff coverage is 8.75%.

@@             Coverage Diff              @@
##           staging     #537       +/-   ##
============================================
- Coverage    92.36%   78.95%   -13.41%     
============================================
  Files            7        7               
  Lines          419      499       +80     
============================================
+ Hits           387      394        +7     
- Misses          32      105       +73     

Impacted Files	Coverage Δ
proglearn/progressive_learner.py	66.52% <8.06%> (-20.48%)	⬇️
proglearn/forest.py	74.19% <11.11%> (-25.81%)	⬇️

---

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 ca15f3b...87fab73. Read the comment docs.

[PSSF23]

You can add unit tests to fix the codecov issue. Also, replicate the change in 9aaa570 to solve the website rendering.

[PSSF23]

You can remove the network.py changes if you don't plan to include the complete network update_task in this PR. I would recommend another PR for that later.

[PSSF23]

@nhahn7 you can include the XOR experiments in this PR. Or you can open another one for those.

[nhahn7]

@PSSF23 Sounds good, I’ll work on adding unit tests and remove the network.py changes today. For the XOR experiments, the notebook currently is just used for plotting (not generating the results). Should I make it in to a tutorial of how to use update_task? Or leave it as is where it just plots from existing results in .txt files?

[PSSF23]

Tutorial format is needed. You can focus on unit tests first.

[nhahn7]

@PSSF23 It looks like its failing due to not recognizing the partial_fit function from sci-kit learn fork. Can this be resolved with changing requirements.txt to specify https://github.com/PSSF23/scikit-learn-stream instead of scikit-learn>=0.22.0?

[PSSF23]

@nhahn7 That would be too much of a dependency. Instead, we can state that the feature is experimental & modify the CircleCI checks.

[PSSF23]

If classes is not currently defined as a variable, I would prefer to have the same name for all classes and input_classes.

[PSSF23]

Update your CircleCI cache from v2 to v3 to resolve the error.

[PSSF23]

As currently there's no way to pass the partial_fit test, just comment out the test for now.

The notebooks and functions should be in experiments. And you did not add the notebook to the rst list.

[PSSF23]

LGTM. We are not merging it to staging as the dependency on SDTF would decrease the code coverage. Hope that this will be resolved in the future.