Investigate build and cicd issues (#206)

* empty change to test build issues

* Correct gensim version

* Correct spaCy version

* Fix several build / dependency bugs

- stop class HeroTypes from inheriting from pd.Series and pd.DataFrame. Inheriting from them is not necessary at all, and causes issues with newer python versions
- skip kmeans doctest as we cannot be sure if it evaluates on [1,0,1,0] or [0,1,0,1] on different builds
- for tests with binary categories (0, 1), test whether either the result or the inversion of the result matches the expected value. This tests the same functionality but prevents failures on different OSes

* Fix check for length of category series before comparison

.travis.yml

@@ -26,4 +26,4 @@ install:
 # 'python3' is a 'command not found' error on Windows but 'py' works on Windows only
 script:
   - black --check . 
-  - python -m unittest discover -s tests -t . || python3 -m unittest discover -s tests -t .
+  - python -m unittest discover -s tests -t . || python3 -m unittest discover -s tests -t .

setup.cfg

@@ -29,13 +29,13 @@ python_requires = >=3.6.1
 install_requires =
     numpy>=1.17
     scikit-learn>=0.22
-    spacy>=2.2.2
+    spacy<3.0.0
     tqdm>=4.3
     nltk>=3.3
     plotly>=4.2.0
     pandas>=1.0.2
     wordcloud>=1.5.0
-    gensim>=3.6.0
+    gensim>=3.6.0,<4.0
     matplotlib>=3.1.0
 # TODO pick the correct version.
 [options.extras_require]

tests/test_representation.py

@@ -222,6 +222,29 @@ def test_dim_reduction_and_clustering_with_vector_series_input(
         else:
             result_s = test_function(vector_s, random_state=42)
 
+        # Binary categories: also test if it equals with
+        # the category labels inverted (e.g. [0, 1, 0] instead
+        # of [1, 0, 1], which makes no difference functionally)
+        if pd.api.types.is_categorical_dtype(result_s):
+            if len(result_s.cat.categories) == 2 and all(
+                result_s.cat.categories == [0, 1]
+            ):
+                try:
+                    result_s_inverted = result_s.apply(lambda category: 1 - category)
+                    pd.testing.assert_series_equal(
+                        s_true,
+                        result_s_inverted,
+                        check_dtype=False,
+                        rtol=0.1,
+                        atol=0.1,
+                        check_category_order=False,
+                        check_categorical=False,
+                    )
+                    return
+                # inverted comparison fails -> continue to normal comparison
+                except AssertionError:
+                    pass
+
         pd.testing.assert_series_equal(
             s_true,
             result_s,
@@ -248,13 +271,37 @@ def test_dim_reduction_and_clustering_with_dataframe_input(
         else:
             result_s = test_function(df, random_state=42)
 
+        # Binary categories: also test if it equals with
+        # the category labels inverted (e.g. [0, 1, 0] instead
+        # of [1, 0, 1], which makes no difference functionally)
+        if pd.api.types.is_categorical_dtype(result_s):
+            if len(result_s.cat.categories) == 2 and all(
+                result_s.cat.categories == [0, 1]
+            ):
+                try:
+                    result_s_inverted = result_s.apply(lambda category: 1 - category)
+                    pd.testing.assert_series_equal(
+                        s_true,
+                        result_s_inverted,
+                        check_dtype=False,
+                        rtol=0.1,
+                        atol=0.1,
+                        check_category_order=False,
+                        check_categorical=False,
+                    )
+                    return
+                # inverted comparison fails -> continue to normal comparison
+                except AssertionError:
+                    pass
+
         pd.testing.assert_series_equal(
             s_true,
             result_s,
             check_dtype=False,
             rtol=0.1,
             atol=0.1,
             check_category_order=False,
+            check_categorical=False,
         )
 
     def test_normalize_DataFrame_also_as_output(self):

texthero/_types.py

@@ -70,7 +70,7 @@ def tfidf(s: TokenSeries) -> DataFrame:
 # This class is mainly for documentation in the docstring.
 
 
-class HeroTypes(pd.Series, pd.DataFrame):
+class HeroTypes:
     """
     Hero Series Types
     =================

texthero/representation.py

@@ -632,7 +632,7 @@ def kmeans(
     >>> s = s.pipe(hero.clean).pipe(hero.tokenize).pipe(
     ...                                         hero.term_frequency
     ...                                             )
-    >>> hero.kmeans(s, n_clusters=2, random_state=42)
+    >>> hero.kmeans(s, n_clusters=2, random_state=42) # doctest: +SKIP
     0    1
     1    0
     2    1