Implement New Single Table DFS Algorithm

[dvreed77]

Pull Request Description

Implement New Single Table DFS Algorithm only using Schema as input

Fixes #2487

[codecov]

Codecov Report

Merging #2516 (255207d) into main (ffaad87) will increase coverage by 0.00%.
The diff coverage is 99.69%.

@@           Coverage Diff            @@
##             main    #2516    +/-   ##
========================================
  Coverage   99.46%   99.47%            
========================================
  Files         392      403    +11     
  Lines       23211    24197   +986     
========================================
+ Hits        23087    24070   +983     
- Misses        124      127     +3     

Impacted Files	Coverage Δ
...ols/tests/testing_utils/generate_fake_dataframe.py	98.00% <98.00%> (ø)
featuretools/feature_discovery/LiteFeature.py	98.95% <98.95%> (ø)
...eaturetools/feature_discovery/FeatureCollection.py	100.00% <100.00%> (ø)
featuretools/feature_discovery/convertors.py	100.00% <100.00%> (ø)
...eaturetools/feature_discovery/feature_discovery.py	100.00% <100.00%> (ø)
featuretools/feature_discovery/type_defs.py	100.00% <100.00%> (ø)
featuretools/feature_discovery/utils.py	100.00% <100.00%> (ø)
featuretools/primitives/utils.py	99.00% <100.00%> (+0.05%)	⬆️
...retools/tests/feature_discovery/test_convertors.py	100.00% <100.00%> (ø)
...tests/feature_discovery/test_feature_collection.py	100.00% <100.00%> (ø)
... and 2 more

[thehomebrewnerd]

I'm not sure if this is the right place for this question, but what happens if we create the f_2 feature which depends on f_1, and then after that f_1 gets renamed (a new alias, I think), does everything still work as expected when it comes time to calculate feature values for f_2? What would happen to the default name of the f_2 feature in this case?

[dvreed77]

Changing the alias has the biggest impact when dealing with origin features and should be avoided as much as possible, but should be fine:

Here is what would happen:

• You create an origin Feature based on column named "Column A"
• You create an engineered Feature using the Absolute primitive to create the name "ABSOLUTE(Column A)"
• You now have 2 features, and if you want to create a feature matrix off of this, you need to convert these back to FeatureBase features first using the conversion function. This will create 2 FeatureBase features with names above and everything should work as expected

Now if you want to change the alias AFTER feature discovery, here is what would happen.

• you set an alias on the origin LiteFeature, to give it a name of "f1", and alter the corresponding column on the dataframe
• you now need to convert the 2 LiteFeatures to FeatureBase features. The Origin Feature would create an IdentityFeature pointing to the "f1" column on the dataframe. Then it would create a TransformFeature, with the name of "ABSOLUTE(Column A)", which probably isn't what you want, but it will still be pointing to the correct base_feature.
• Essentially everything works as expected, its just the names won't be right.