Multiple output column features with primitive refactor #376

rwedge · 2019-01-16T21:11:05Z

This pull request updates the API for handling features that output multiple columns of values. In the past these were referred to in the code as "expanding" features. A brief rundown of the changes:

Multiple columns in the feature matrix can now correspond to the same feature. Where before a feature with multiple outputs would have a list of values per instance, these features now get one column per value.
Column names do not always match a feature name. With features having multiple columns it's necessary to modify the column names to distinguish between the various output columns of a feature.
feature.expanding (bool) replaced with feature.primitive.number_output_features (int). Instead of a feature either being expanding or not, it has an integer corresponding to the number of output features (columns) it has.
feature.get_feature_names() is the correct way to get the names of columns in a feature matrix
feature.default_value is a singular value that becomes the default for all output features. Instead of a list of default values, one for each output feature, each output features shares the same default value.
new keyword argument for make_primitive functions
- number_output_features (int): sets the number of output features for a custom primitive
features with multiple outputs can now be used as direct features

* remove incorrect commutative attributes * handle rsub override and test reverse overrides * rename weekend primitive is_weekend * updated weekend to is_weekend in docs * test values for scalar_subtract_numeric * rename subtract_numeric and scalar_subtract_numeric to subtract_numeric_feature and scalar_subtract_numeric_feature * revert subtract_numeric_feature to subtract_numeric

codecov · 2019-01-16T21:22:21Z

Codecov Report

Merging #376 into master will increase coverage by 0.21%.
The diff coverage is 98.8%.

@@            Coverage Diff             @@
##           master     #376      +/-   ##
==========================================
+ Coverage   95.56%   95.78%   +0.21%     
==========================================
  Files          89       89              
  Lines        7555     7730     +175     
==========================================
+ Hits         7220     7404     +184     
+ Misses        335      326       -9

Impacted Files	Coverage Δ
featuretools/synthesis/deep_feature_synthesis.py	`94.49% <100%> (+0.96%)`	⬆️
...aturetools/tests/primitive_tests/test_agg_feats.py	`99.11% <100%> (+0.07%)`	⬆️
...etools/primitives/base/transform_primitive_base.py	`100% <100%> (ø)`	⬆️
featuretools/feature_base/feature_base.py	`95.81% <100%> (+0.77%)`	⬆️
...utational_backend/test_calculate_feature_matrix.py	`99.3% <100%> (ø)`	⬆️
featuretools/primitives/base/primitive_base.py	`100% <100%> (ø)`	⬆️
...ols/tests/dfs_tests/test_deep_feature_synthesis.py	`98.32% <100%> (+0.05%)`	⬆️
...ools/primitives/base/aggregation_primitive_base.py	`100% <100%> (ø)`	⬆️
...ests/computational_backend/test_encode_features.py	`100% <100%> (ø)`	⬆️
...ools/primitives/standard/aggregation_primitives.py	`95.45% <100%> (+0.34%)`	⬆️
... and 6 more

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 36ce3c3...e5d253c. Read the comment docs.

kmax12 · 2019-01-16T22:07:42Z

featuretools/feature_base/feature_base.py

        self.base_feature = base_feature
+        self.primitive = base_feature.primitive


hmm, not sure about this. i would this our code shouldn't ever try access the primitive property of a direct feature

kmax12 · 2019-01-16T22:11:37Z

featuretools/primitives/standard/aggregation_primitives.py

+        return np.nan
+
+    def get_function(self):
+        def pd_topn(x, n=self.number_output_features):


i think we can rename pd_topn to just n_most_common

kmax12 · 2019-01-16T22:12:06Z

featuretools/primitives/standard/aggregation_primitives.py

+        self.number_output_features = n
+
+    @property
+    def default_value(self):


do we need this? i think primitive base already has default value as np.nan

kmax12 · 2019-01-16T22:13:09Z

featuretools/primitives/standard/aggregation_primitives.py

+    name = "n_most_common"
+    input_types = [Discrete]
+    return_type = Discrete
+    stack_on = []


i think we want to let NMostCommon stack

kmax12 · 2019-01-16T22:14:02Z

featuretools/primitives/standard/aggregation_primitives.py

+        def pd_topn(x, n=self.number_output_features):
+            array = np.array(x.value_counts()[:n].index)
+            if len(array) < n:
+                filler = np.full(n - len(array), self.default_value)


let's not use self.default_value here

kmax12 · 2019-01-16T22:35:01Z

featuretools/computational_backends/pandas_backend.py

+            if f.primitive.number_output_features > 1:
+                names = f.get_feature_names()
+                assert len(names) == len(values)
+                d.update(dict(zip(names, values)))


any reason use use DataFrame.update here instead of the loop approach from _calculate_transform_features?

for name, value in zip(names, values): frame[name] = strip_values_if_series(value)

featuretools/synthesis/deep_feature_synthesis.py

kmax12 · 2019-01-16T22:45:02Z

featuretools/tests/computational_backend/test_pandas_backend.py

@@ -499,3 +524,8 @@ def test_empty_child_dataframe():
    fm2 = ft.calculate_feature_matrix(entityset=es, features=[count_where, trend_where], cutoff_time=pd.Timestamp("1/4/2018"))
    names = [count_where.get_name(), trend_where.get_name()]
    assert_array_equal(fm2[names], [[0, np.nan]])
+
+
+def test_computes_direct_features_of_multi_ouput_features():


let's create an issue for this

this got implemented in test_direct_features.py so I'll remove it here

kmax12 · 2019-01-16T22:49:38Z

featuretools/tests/primitive_tests/test_direct_features.py

@@ -44,3 +58,66 @@ def test_direct_rename(es):
    assert feat.get_name() != copy_feat.get_name()
    assert feat.base_features[0].generate_name() == copy_feat.base_features[0].generate_name()
    assert feat.entity == copy_feat.entity
+
+
+def test_direct_of_multi_output_transform_feat(es):


maybe add that this is just testing naming

kmax12 · 2019-01-16T22:49:59Z

featuretools/tests/primitive_tests/test_direct_features.py

+    fm, fl = dfs(
+        entityset=es,
+        target_entity="sessions",
+        trans_primitives=[TestTime, Year, Month, Day, Hour, Minute, Second])


i think you can do features_only=True here. same for test case below

…s a primitive

kmax12 and others added 30 commits December 31, 2018 17:52

initial support

06256c6

cfm tests passing

a0cd617

linting

2de087c

progress

d0e786d

progress

d36701d

add lint-fix and autopep

e03bf2d

more passing

352ff58

progress

133e888

progress

aea1236

progress

e592ad3

don't require initialization

f2dab2f

separate override tests

33fe750

lint

c37161c

cleanup

ed11264

more tests passing

4674174

one more test passing

0765d96

all but one test passing

fc68f4b

all test pass

8c8d73d

linting

83f63ad

add to lint-fix

cec3928

all tests pass!

67b0244

cleanup

1567b2c

fix tests

aad7553

fix test

2bd20f7

fix linting

42945aa

docs building

869c8a8

actually fix docs

6e6ead3

updates

39ad144

update rsub

91eda86

rwedge and others added 5 commits January 16, 2019 12:30

remove get_feature_names on primitive

5557c54

cleanup

794b2c8

test direct features of multi output features

c6f3d0d

Merge branch 'primitive-refactor' into multi-output-primitive-refactor

834382f

linting

cc69126

rwedge mentioned this pull request Jan 16, 2019

[WIP] Primitives with multiple columns #350

Closed

kmax12 suggested changes Jan 16, 2019

View reviewed changes

rwedge and others added 17 commits January 17, 2019 16:29

remove unecessary primitive initializations

05e3183

rename NMostCommon primitive function

86431df

remove unecessary default value from NMostCommon

ca22c74

allow NMostCommon to stack

02e7882

don't set primitive for DirectFeature

19dfb3c

make number_output_features a property of FeatureBase

84118fe

change parameter name in check_stacking

43bb673

rename ready_primitive to handle_primitive and add check that input i…

d305d34

…s a primitive

update variable name in where primitive check

1d063cb

switch to using feature.number_output_features in logic

4965121

add warning in encode_features for multi output features

85e8457

make helper function for adding feature values to existing data

fb36e56

Merge branch 'primitive-refactor' into multi-output-primitive-refactor

ca7bf16

clean up direct feature tests on multi-output features

fdb3092

addressed review comments

40946ba

Merge branch 'primitive-refactor' into multi-output-primitive-refactor

2a41510

remove logic to do direct features of grandparents

f4bb67a

rwedge changed the base branch from primitive-refactor to master January 18, 2019 23:39

Merge branch 'master' into multi-output-primitive-refactor

e5d253c

rwedge merged commit f80b5ae into master Jan 19, 2019

rwedge mentioned this pull request Jan 30, 2019

v0.6.0 #387

Merged

rwedge deleted the multi-output-primitive-refactor branch February 19, 2021 21:24

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Multiple output column features with primitive refactor #376

Multiple output column features with primitive refactor #376

rwedge commented Jan 16, 2019 •

edited

Loading

codecov bot commented Jan 16, 2019 •

edited

Loading

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

rwedge Jan 17, 2019 •

edited

Loading

kmax12 Jan 16, 2019

kmax12 Jan 16, 2019

		self.base_feature = base_feature
		self.primitive = base_feature.primitive

Multiple output column features with primitive refactor #376

Multiple output column features with primitive refactor #376

Conversation

rwedge commented Jan 16, 2019 • edited Loading

codecov bot commented Jan 16, 2019 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rwedge Jan 17, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rwedge commented Jan 16, 2019 •

edited

Loading

codecov bot commented Jan 16, 2019 •

edited

Loading

rwedge Jan 17, 2019 •

edited

Loading