127 adjust threshold

[yiwen-h]

• When training models, dataset now split into train, validation, test
• Validation set used to calculate thresholds for best F1 scores for each label
• Calculated thresholds included in Excel analysis document for each model
• predict_multilabel now takes as argument a dict containing thresholds to use for each label

Testing has shown small improvements to performance
Next steps: integrate into the API

[codecov]

Codecov Report

Patch coverage: 81.35% and project coverage change: +0.41% 🎉

Comparison is base (71ece91) 88.36% compared to head (5c1c217) 88.78%.
Report is 16 commits behind head on development.

Additional details and impacted files

@@               Coverage Diff               @@
##           development     #136      +/-   ##
===============================================
+ Coverage        88.36%   88.78%   +0.41%     
===============================================
  Files               11       11              
  Lines              937      945       +8     
===============================================
+ Hits               828      839      +11     
+ Misses             109      106       -3     

Files Changed	Coverage Δ
pxtextmining/pipelines/multilabel_pipeline.py	88.54% <67.64%> (-11.46%)	⬇️
...ining/factories/factory_predict_unlabelled_text.py	83.16% <85.48%> (-0.36%)	⬇️
pxtextmining/factories/factory_write_results.py	75.00% <85.71%> (-1.37%)	⬇️
...xtextmining/factories/factory_model_performance.py	93.69% <92.30%> (+16.01%)	⬆️
pxtextmining/factories/factory_pipeline.py	92.94% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[yiwen-h]

closes #127

[ChrisBeeley]

👍