Polars feature updates (#1119)

* Add polars to datetime_column_profile

* Polars added to unstructured labeler

* Quick fix for keras and tensorflow

* Add polars to unstructured text

* Correct polars usage

* Dask version

* Quick fix for keras and tensorflow

* Minor polars updates

* Change type for isinstance

dataprofiler/profilers/datetime_column_profile.py

@@ -7,6 +7,7 @@
 
 import numpy as np
 import pandas as pd
+import polars as pl
 
 from . import profiler_utils
 from .base_column_profilers import BaseColumnPrimitiveTypeProfiler, BaseColumnProfiler
@@ -256,8 +257,7 @@ def _get_datetime_profile(cls, df_series: pd.Series) -> dict:
         profile: dict = dict()
         activated_date_formats: list = list()
         len_df = len(df_series)
-
-        is_row_datetime = pd.Series(np.full((len(df_series)), False))
+        is_row_datetime = pd.Series(np.full((len_df), False))
 
         min_value = None
         max_value = None
@@ -275,18 +275,19 @@ def _get_datetime_profile(cls, df_series: pd.Series) -> dict:
                 )
             )
 
-            df_dates = valid_dates[~valid_dates.isnull()]
+            df_dates = pl.Series(valid_dates[~valid_dates.isnull()])
 
-            if "%b" in date_format and not df_dates.empty:
+            if "%b" in date_format and not df_dates.is_empty():
                 may_month = 5  # May can be %b or %B we want to force, so check
-                all_may = df_dates.apply(lambda x: x.month == may_month).all()
+                all_may = df_dates.map_elements(lambda x: x.month == may_month)
+                all_may = pl.Series(all_may).all()
                 if all_may:
-                    valid_dates[:] = np.nan
-                    df_dates = pd.Series([], dtype=object)
+                    valid_dates[:] = None
+                    df_dates = pl.Series([])
 
             # Create mask to avoid null dates
             null_date_mask = valid_dates.isnull()
-            np_date_array = df_dates.values
+            np_date_array = df_dates.to_numpy()
 
             # check off any values which were found to be datetime
             is_row_datetime[~is_row_datetime] = (~null_date_mask).values
@@ -298,18 +299,18 @@ def _get_datetime_profile(cls, df_series: pd.Series) -> dict:
                 max_idx = np.argmax(np_date_array)
 
                 # Selects the min, ma value objects for comparison
-                tmp_min_value_obj = df_dates.iloc[min_idx]
-                tmp_max_value_obj = df_dates.iloc[max_idx]
+                tmp_min_value_obj = df_dates.item(int(min_idx))
+                tmp_max_value_obj = df_dates.item(int(max_idx))
 
                 # If minimum value, keep reference
                 if tmp_min_value_obj < min_value_obj:
                     min_value = df_series[~null_date_mask].iloc[min_idx]
-                    min_value_obj = tmp_min_value_obj
+                    min_value_obj = pd.Timestamp(tmp_min_value_obj)
 
                 # If maximum value, keep reference
                 if tmp_max_value_obj > max_value_obj:
                     max_value = df_series[~null_date_mask].iloc[max_idx]
-                    max_value_obj = tmp_max_value_obj
+                    max_value_obj = pd.Timestamp(tmp_max_value_obj)
 
             df_series = df_series[null_date_mask]
 

dataprofiler/profilers/profiler_utils.py

@@ -752,7 +752,7 @@ def perform_chi_squared_test_for_homogeneity(
     # If one or less categories, we have zero/negative degrees of freedom,
     # which is not an appropriate value for this context
     num_cats = len(cat_counts)
-    if len(cat_counts) <= 1:
+    if num_cats <= 1:
         warnings.warn(
             "Insufficient number of categories. "
             "Chi-squared test cannot be performed.",

dataprofiler/profilers/unstructured_labeler_profile.py

@@ -3,7 +3,7 @@
 
 from collections import defaultdict
 
-from pandas import Series
+import polars as pl
 
 from ..labelers.base_data_labeler import BaseDataLabeler
 from ..labelers.data_labelers import DataLabeler
@@ -155,25 +155,27 @@ def label_encoding(self) -> list[str]:
         return self.data_labeler.labels
 
     @BaseColumnProfiler._timeit(name="data_labeler_predict")
-    def _update_helper(self, df_series_clean: Series, profile: dict) -> None:
+    def _update_helper(self, df_series_clean: pl.Series, profile: dict) -> None:
         """
         Update col profile properties with clean dataset and its known profile.
 
         :param df_series_clean: df series with nulls removed
-        :type df_series_clean: pandas.core.series.Series
+        :type df_series_clean: polars.Series
         :param profile: profile dictionary
         :type profile: dict
         :return: None
         """
+        data_ndarray = df_series_clean.to_numpy()
+
         # this will get char_level predictions as output
-        predictions = self.data_labeler.predict(df_series_clean)
+        predictions = self.data_labeler.predict(data_ndarray)
 
         # also store spacy/NER format
         postprocessor = CharPostprocessor(
             use_word_level_argmax=True, output_format="NER"
         )
         format_predictions = postprocessor.process(
-            df_series_clean, predictions.copy(), self.data_labeler.label_mapping
+            data_ndarray, predictions.copy(), self.data_labeler.label_mapping
         )
 
         # Update counts and percent values
@@ -188,14 +190,15 @@ def _update_helper(self, df_series_clean: Series, profile: dict) -> None:
         # CHARACTERS/WORDS PROCESSED
         self._update_column_base_properties(profile)
 
-    def update(self, df_series: Series) -> None:
+    def update(self, df_series: pl.Series) -> None:
         """Update profile."""
         if len(df_series) == 0:
             return
         profile = dict(
             char_sample_size=self.char_sample_size,
             word_sample_size=self.word_sample_size,
         )
+
         self._update_helper(df_series, profile)
 
     @property
@@ -278,21 +281,21 @@ def _update_true_char_label_counts(self, predictions: list) -> None:
             self.char_sample_size += len(sample)
 
     def _update_postprocess_char_label_counts(
-        self, df_series_clean: Series, format_predictions: dict
+        self, df_series_clean: pl.Series, format_predictions: dict
     ) -> None:
         """
         Update the postprocess character label counts.
 
         :param df_series_clean: df series with nulls removed
-        :type df_series_clean: pandas.core.series.Series
+        :type df_series_clean: polars.Series
         :param format_predictions: contains dict of samples with predictions on
             the character level in congruence with the word level predictions
         :type format_predictions: Dict
         :return: None
         """
         char_label_counts = self.entity_counts["postprocess_char_level"]
 
-        for index, result in enumerate(zip(df_series_clean, format_predictions)):
+        for result in zip(df_series_clean, format_predictions):
             text, entities = result
             index = 0
             for entity in entities:
@@ -308,20 +311,20 @@ def _update_postprocess_char_label_counts(
             char_label_counts["UNKNOWN"] += len(text) - index
 
     def _update_word_label_counts(
-        self, df_series_clean: Series, format_predictions: dict
+        self, df_series_clean: pl.Series, format_predictions: dict
     ) -> None:
         """
         Update the sorted dictionary of each entity count.
 
         :param df_series_clean: df series with nulls removed
-        :type df_series_clean: pandas.core.series.Series
+        :type df_series_clean: polars.Series
         :param format_predictions: Dictionary of sample text and entities
         :type format_predictions: dict
         :return: None
         """
         word_label_counts = self.entity_counts["word_level"]
 
-        for index, result in enumerate(zip(df_series_clean, format_predictions)):
+        for result in zip(df_series_clean, format_predictions):
             text, entities = result
             begin_word_idx = -1
             index = 0

dataprofiler/profilers/unstructured_text_profile.py

@@ -6,6 +6,7 @@
 import warnings
 from collections import Counter, defaultdict
 
+import polars as pl
 from numpy import ndarray
 from pandas import DataFrame, Series
 
@@ -667,15 +668,15 @@ def profile(self) -> dict:
     @BaseColumnProfiler._timeit(name="vocab")
     def _update_vocab(
         self,
-        data: list | ndarray | DataFrame,
+        data: list | ndarray | DataFrame | pl.DataFrame,
         prev_dependent_properties: dict = None,
         subset_properties: dict = None,
     ) -> None:
         """
         Find the vocabulary counts used in the text samples.
 
         :param data: list or array of data from which to extract vocab
-        :type data: Union[list, numpy.array, pandas.DataFrame]
+        :type data: Union[list, numpy.array, pandas.DataFrame, polars.DataFrame]
         :param prev_dependent_properties: Contains all the previous properties
             that the calculations depend on.
         :type prev_dependent_properties: dict
@@ -690,15 +691,15 @@ def _update_vocab(
     @BaseColumnProfiler._timeit(name="words")
     def _update_words(
         self,
-        data: list | ndarray | DataFrame,
+        data: list | ndarray | DataFrame | pl.DataFrame,
         prev_dependent_properties: dict = None,
         subset_properties: dict = None,
     ) -> None:
         """
         Find unique words and word count used in the text samples.
 
         :param data: list or array of data from which to extract vocab
-        :type data: Union[list, numpy.array, pandas.DataFrame]
+        :type data: Union[list, numpy.array, pandas.DataFrame, polars.DataFrame]
         :param prev_dependent_properties: Contains all the previous properties
             that the calculations depend on.
         :type prev_dependent_properties: dict
@@ -708,37 +709,54 @@ def _update_words(
         :return: None
         """
         if not self._is_case_sensitive:
-            words = (
-                [w.strip(string.punctuation) for w in row.lower().split()]
-                for row in data
-            )
+            if isinstance(data, pl.DataFrame):
+                words = (
+                    [
+                        w.strip(string.punctuation)
+                        for w in row.str.to_lowercase().str.split(by=" ")
+                    ]
+                    for row in data
+                )
+            else:
+                words = (
+                    [w.strip(string.punctuation) for w in row.lower().split()]
+                    for row in data
+                )
         else:
-            words = ([w.strip(string.punctuation) for w in row.split()] for row in data)
+            if isinstance(data, pl.DataFrame):
+                words = (
+                    [w.strip(string.punctuation) for w in row.str.split(by=" ")]
+                    for row in data
+                )
+            else:
+                words = (
+                    [w.strip(string.punctuation) for w in row.split()] for row in data
+                )
         word_count = Counter(itertools.chain.from_iterable(words))
 
         for w, c in word_count.items():
             if w and w.lower() not in self._stop_words:
                 self.word_count.update({w: c})
 
-    def _update_helper(self, data: Series, profile: dict) -> None:
+    def _update_helper(self, data: pl.Series, profile: dict) -> None:
         """
         Update col profile properties with clean dataset and its known null parameters.
 
         :param data: df series with nulls removed
-        :type data: pandas.core.series.Series
+        :type data: polars.Series
         :param profile: text profile dictionary
         :type profile: dict
         :return: None
         """
         self.sample_size += profile.pop("sample_size")
         self.metadata = profile
 
-    def update(self, data: Series) -> TextProfiler:
+    def update(self, data: Series | pl.Series) -> TextProfiler:
         """
         Update the column profile.
 
         :param data: df series
-        :type data: pandas.core.series.Series
+        :type data: polars.Series
         :return: updated TextProfiler
         :rtype: TextProfiler
         """
@@ -748,14 +766,19 @@ def update(self, data: Series) -> TextProfiler:
 
         profile = dict(sample_size=len_data)
 
+        if isinstance(data, pl.Series):
+            data_pandas = data.to_pandas()
+        else:
+            data_pandas = data
+
         BaseColumnProfiler._perform_property_calcs(
             self,  # type: ignore
             self.__calculations,
-            df_series=data,
+            df_series=data_pandas,
             prev_dependent_properties={},
             subset_properties=profile,
         )
 
-        self._update_helper(data, profile)
+        self._update_helper(pl.Series(data), profile)
 
         return self