Change HP Name & Include Text example

autosklearn/pipeline/components/data_preprocessing/text_encoding/bag_of_word_encoding.py

@@ -17,13 +17,13 @@
 class BagOfWordEncoder(AutoSklearnPreprocessingAlgorithm):
     def __init__(
         self,
-        ngram_range: int = 1,
+        ngram_upper_bound: int = 1,
         min_df_choice: str = "min_df_absolute",
         min_df_absolute: int = 0,
         min_df_relative: float = 0.01,
         random_state: Optional[Union[int, np.random.RandomState]] = None,
     ) -> None:
-        self.ngram_range = ngram_range
+        self.ngram_upper_bound = ngram_upper_bound
         self.random_state = random_state
         self.min_df_choice = min_df_choice
         self.min_df_absolute = min_df_absolute
@@ -46,13 +46,13 @@ def fit(
         if self.min_df_choice == "min_df_absolute":
             self.preprocessor = CountVectorizer(
                 min_df=self.min_df_absolute,
-                ngram_range=(1, self.ngram_range),
+                ngram_range=(1, self.ngram_upper_bound),
             )
 
         elif self.min_df_choice == "min_df_relative":
             self.preprocessor = CountVectorizer(
                 min_df=self.min_df_relative,
-                ngram_range=(1, self.ngram_range),
+                ngram_range=(1, self.ngram_upper_bound),
             )
 
         else:
@@ -98,8 +98,8 @@ def get_hyperparameter_search_space(
         dataset_properties: Optional[DATASET_PROPERTIES_TYPE] = None,
     ) -> ConfigurationSpace:
         cs = ConfigurationSpace()
-        hp_ngram_range = CSH.UniformIntegerHyperparameter(
-            name="ngram_range", lower=1, upper=3, default_value=1
+        hp_ngram_upper_bound = CSH.UniformIntegerHyperparameter(
+            name="ngram_upper_bound", lower=1, upper=3, default_value=1
         )
         hp_min_df_choice_bow = CSH.CategoricalHyperparameter(
             "min_df_choice", choices=["min_df_absolute", "min_df_relative"]
@@ -112,7 +112,7 @@ def get_hyperparameter_search_space(
         )
         cs.add_hyperparameters(
             [
-                hp_ngram_range,
+                hp_ngram_upper_bound,
                 hp_min_df_choice_bow,
                 hp_min_df_absolute_bow,
                 hp_min_df_relative_bow,

autosklearn/pipeline/components/data_preprocessing/text_encoding/bag_of_word_encoding_distinct.py

@@ -16,13 +16,13 @@
 class BagOfWordEncoder(AutoSklearnPreprocessingAlgorithm):
     def __init__(
         self,
-        ngram_range: int = 1,
+        ngram_upper_bound: int = 1,
         min_df_choice: str = "min_df_absolute",
         min_df_absolute: int = 0,
         min_df_relative: float = 0.01,
         random_state: Optional[Union[int, np.random.RandomState]] = None,
     ) -> None:
-        self.ngram_range = ngram_range
+        self.ngram_upper_bound = ngram_upper_bound
         self.random_state = random_state
         self.min_df_choice = min_df_choice
         self.min_df_absolute = min_df_absolute
@@ -40,7 +40,8 @@ def fit(
 
                 for feature in X.columns:
                     vectorizer = CountVectorizer(
-                        min_df=self.min_df_absolute, ngram_range=(1, self.ngram_range)
+                        min_df=self.min_df_absolute,
+                        ngram_range=(1, self.ngram_upper_bound),
                     ).fit(X[feature])
                     self.preprocessor[feature] = vectorizer
 
@@ -50,7 +51,8 @@ def fit(
 
                 for feature in X.columns:
                     vectorizer = CountVectorizer(
-                        min_df=self.min_df_relative, ngram_range=(1, self.ngram_range)
+                        min_df=self.min_df_relative,
+                        ngram_range=(1, self.ngram_upper_bound),
                     ).fit(X[feature])
                     self.preprocessor[feature] = vectorizer
             else:
@@ -102,8 +104,8 @@ def get_hyperparameter_search_space(
         dataset_properties: Optional[DATASET_PROPERTIES_TYPE] = None,
     ) -> ConfigurationSpace:
         cs = ConfigurationSpace()
-        hp_ngram_range = CSH.UniformIntegerHyperparameter(
-            name="ngram_range", lower=1, upper=3, default_value=1
+        hp_ngram_upper_bound = CSH.UniformIntegerHyperparameter(
+            name="ngram_upper_bound", lower=1, upper=3, default_value=1
         )
         hp_min_df_choice_bow = CSH.CategoricalHyperparameter(
             "min_df_choice", choices=["min_df_absolute", "min_df_relative"]
@@ -116,7 +118,7 @@ def get_hyperparameter_search_space(
         )
         cs.add_hyperparameters(
             [
-                hp_ngram_range,
+                hp_ngram_upper_bound,
                 hp_min_df_choice_bow,
                 hp_min_df_absolute_bow,
                 hp_min_df_relative_bow,

autosklearn/pipeline/components/data_preprocessing/text_encoding/tfidf_encoding.py

@@ -17,14 +17,14 @@
 class TfidfEncoder(AutoSklearnPreprocessingAlgorithm):
     def __init__(
         self,
-        ngram_range: int = 1,
+        ngram_upper_bound: int = 1,
         use_idf: bool = True,
         min_df_choice: str = "min_df_absolute",
         min_df_absolute: int = 0,
         min_df_relative: float = 0.01,
         random_state: Optional[Union[int, np.random.RandomState]] = None,
     ) -> None:
-        self.ngram_range = ngram_range
+        self.ngram_upper_bound = ngram_upper_bound
         self.random_state = random_state
         self.use_idf = use_idf
         self.min_df_choice = min_df_choice
@@ -50,14 +50,14 @@ def fit(
             self.preprocessor = TfidfVectorizer(
                 min_df=self.min_df_absolute,
                 use_idf=self.use_idf,
-                ngram_range=(1, self.ngram_range),
+                ngram_range=(1, self.ngram_upper_bound),
             )
 
         elif self.min_df_choice == "min_df_relative":
             self.preprocessor = TfidfVectorizer(
                 min_df=self.min_df_relative,
                 use_idf=self.use_idf,
-                ngram_range=(1, self.ngram_range),
+                ngram_range=(1, self.ngram_upper_bound),
             )
 
         else:
@@ -103,8 +103,8 @@ def get_hyperparameter_search_space(
         dataset_properties: Optional[DATASET_PROPERTIES_TYPE] = None,
     ) -> ConfigurationSpace:
         cs = ConfigurationSpace()
-        hp_ngram_range = CSH.UniformIntegerHyperparameter(
-            name="ngram_range", lower=1, upper=3, default_value=1
+        hp_ngram_upper_bound = CSH.UniformIntegerHyperparameter(
+            name="ngram_upper_bound", lower=1, upper=3, default_value=1
         )
         hp_use_idf = CSH.CategoricalHyperparameter("use_idf", choices=[False, True])
         hp_min_df_choice = CSH.CategoricalHyperparameter(
@@ -118,7 +118,7 @@ def get_hyperparameter_search_space(
         )
         cs.add_hyperparameters(
             [
-                hp_ngram_range,
+                hp_ngram_upper_bound,
                 hp_use_idf,
                 hp_min_df_choice,
                 hp_min_df_absolute,

examples/40_advanced/example_text_preprocessing.py

@@ -1,79 +1,104 @@
 # -*- encoding: utf-8 -*-
 """
 ==================
-Text Preprocessing
+Text preprocessing
 ==================
-This example shows, how to use text features in *auto-sklearn*. *auto-sklearn* can automatically
-encode text features if they are provided as string type in a pandas dataframe.
 
-For processing text features you need a pandas dataframe and set the desired
-text columns to string and the categorical columns to category.
+The following example shows how to fit a simple NLP problem with
+*auto-sklearn*.
 
-*auto-sklearn* text embedding creates a bag of words count.
+For an introduction to text preprocessing you can follow these links:
+    1. https://scikit-learn.org/stable/tutorial/text_analytics/working_with_text_data.html
+    2. https://machinelearningmastery.com/clean-text-machine-learning-python/
 """
+from pprint import pprint
+
+import pandas as pd
 import sklearn.metrics
-import sklearn.datasets
+from sklearn.datasets import fetch_20newsgroups
+
 import autosklearn.classification
 
 ############################################################################
 # Data Loading
 # ============
 
-X, y = sklearn.datasets.fetch_openml(data_id=40945, return_X_y=True)
-
-# by default, the columns which should be strings are not formatted as such
-print(f"{X.info()}\n")
-
-# manually convert these to string columns
-X = X.astype(
-    {
-        "name": "string",
-        "ticket": "string",
-        "cabin": "string",
-        "boat": "string",
-        "home.dest": "string",
-    }
-)
-
-# now *auto-sklearn* handles the string columns with its text feature preprocessing pipeline
-
-X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(
-    X, y, random_state=1
-)
+newsgroups_train = fetch_20newsgroups(subset="train", random_state=42, shuffle=True)
+newsgroups_test = fetch_20newsgroups(subset="test")
+
+# load train data
+df_train = pd.DataFrame({"X": [], "y": []})
+
+for idx, (text, target) in enumerate(
+    zip(newsgroups_train.data, newsgroups_train.target)
+):
+    df_train = pd.concat(
+        [
+            df_train,
+            pd.DataFrame(
+                {"X": text, "y": newsgroups_train.target_names[target]}, index=[idx]
+            ),
+        ]
+    )
+
+# explicitly label text column as string
+X_train = df_train.astype({"X": "string", "y": "category"})
+
+# show all 20 labels
+print(list(newsgroups_train.target_names))
+
+# reduce the example to only 5 labels
+five_newsgroups_labels = list(newsgroups_train.target_names)[:5]
+
+X_train = X_train[~X_train["y"].isin(five_newsgroups_labels)]
+y_train = X_train.pop("y")
+
+# load test data
+df_test = pd.DataFrame({"X": [], "y": []})
+
+for idx, (text, target) in enumerate(zip(newsgroups_test.data, newsgroups_test.target)):
+    df_test = pd.concat(
+        [
+            df_train,
+            pd.DataFrame(
+                {"X": text, "y": newsgroups_train.target_names[int(target)]},
+                index=[idx],
+            ),
+        ]
+    )
+
+# explicitly label text column as string
+X_test = df_test.astype({"X": "string", "y": "category"})
+X_test = X_test[~X_test["y"].isin(five_newsgroups_labels)]
+y_test = X_test.pop("y")
 
-cls = autosklearn.classification.AutoSklearnClassifier(
-    time_left_for_this_task=30,
-    # Bellow two flags are provided to speed up calculations
-    # Not recommended for a real implementation
-    initial_configurations_via_metalearning=0,
-    smac_scenario_args={"runcount_limit": 1},
+############################################################################
+# Build and fit a classifier
+# ==========================
+
+automl = autosklearn.classification.AutoSklearnClassifier(
+    # set the time high enough text preprocessing can create many new features
+    time_left_for_this_task=300,
+    per_run_time_limit=30,
+    tmp_folder="/tmp/autosklearn_text_example_tmp",
 )
+automl.fit(X_train, y_train, dataset_name="20_Newsgroups")
 
-cls.fit(X_train, y_train, X_test, y_test)
-
-predictions = cls.predict(X_test)
-print("Accuracy score", sklearn.metrics.accuracy_score(y_test, predictions))
-
+############################################################################
+# View the models found by auto-sklearn
+# =====================================
 
-X, y = sklearn.datasets.fetch_openml(data_id=40945, return_X_y=True, as_frame=True)
-X = X.select_dtypes(exclude=["object"])
+print(automl.leaderboard())
 
-X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(
-    X, y, random_state=1
-)
+############################################################################
+# Print the final ensemble constructed by auto-sklearn
+# ====================================================
 
-cls = autosklearn.classification.AutoSklearnClassifier(
-    time_left_for_this_task=30,
-    # Bellow two flags are provided to speed up calculations
-    # Not recommended for a real implementation
-    initial_configurations_via_metalearning=0,
-    smac_scenario_args={"runcount_limit": 1},
-)
+pprint(automl.show_models(), indent=4)
 
-cls.fit(X_train, y_train, X_test, y_test)
+###########################################################################
+# Get the Score of the final ensemble
+# ===================================
 
-predictions = cls.predict(X_test)
-print(
-    "Accuracy score without text preprocessing",
-    sklearn.metrics.accuracy_score(y_test, predictions),
-)
+predictions = automl.predict(X_test)
+print("Accuracy score:", sklearn.metrics.accuracy_score(y_test, predictions))

test/test_pipeline/components/data_preprocessing/test_data_preprocessing_text.py

@@ -20,7 +20,7 @@ def test_fit_transform(self):
             }
         ).astype({"col1": "string", "col2": "string"})
         BOW_fitted = BOW(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -46,7 +46,7 @@ def test_fit_transform(self):
         np.testing.assert_array_equal(Yt, Y)
 
         BOW_fitted = BOW_distinct(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -69,7 +69,7 @@ def test_transform(self):
             }
         ).astype({"col1": "string", "col2": "string"})
         X_t = BOW(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -81,7 +81,7 @@ def test_transform(self):
         np.testing.assert_array_equal(X_t.toarray(), y)
 
         X_t = BOW_distinct(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -103,7 +103,7 @@ def test_check_shape(self):
             }
         ).astype({"col1": "string", "col2": "string"})
         X_t = BOW(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -113,7 +113,7 @@ def test_check_shape(self):
         self.assertEqual(X_t.shape, (2, 5))
 
         X_t = BOW_distinct(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -130,7 +130,7 @@ def test_check_nan(self):
             }
         ).astype({"col1": "string", "col2": "string"})
         X_t = BOW(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,
@@ -140,7 +140,7 @@ def test_check_nan(self):
         self.assertEqual(X_t.shape, (3, 5))
 
         X_t = BOW_distinct(
-            ngram_range=1,
+            ngram_upper_bound=1,
             min_df_choice="min_df_absolute",
             min_df_absolute=0,
             min_df_relative=0,