Revert "Merge pull request #426 from cisco/vidamoda/update-sklearn"

This reverts commit f158106, reversing
changes made to 76963ea.

A bug fix for the 4.6.0 release was inadvertently included in a subsequent PR.
This reverts PR#426 so that the fix can be added as 4.6.1 without including
the rest of PR#426

mindmeld/components/_config.py

@@ -783,8 +783,8 @@ def get_classifier_config(
                                    copy.deepcopy(getattr(module_conf, attr_name)))
     except AttributeError:
         try:
-            result = merge_param_configs(_get_default_classifier_config(clf_type), copy.deepcopy(
-                getattr(module_conf, CONFIG_DEPRECATION_MAPPING[attr_name]))
+            result = copy.deepcopy(
+                getattr(module_conf, CONFIG_DEPRECATION_MAPPING[attr_name])
             )
             msg = (
                 "%s config is deprecated. Please use the equivalent %s config "

mindmeld/components/classifier.py

@@ -403,8 +403,6 @@ def _get_model_config(loaded_config=None, **kwargs):
         Returns:
             ModelConfig: The model configuration corresponding to the provided config name
         """
-        if 'params' in loaded_config and 'params' in kwargs:
-            kwargs['params'] = {**loaded_config['params'], **kwargs['params']}
         try:
             # If all params required for model config were passed in, use kwargs
             return ModelConfig(**kwargs)
@@ -413,6 +411,9 @@ def _get_model_config(loaded_config=None, **kwargs):
             if not loaded_config:
                 logger.warning("loaded_config is not passed in")
             model_config = loaded_config or {}
+            if 'params' in model_config and 'params' in kwargs:
+                kwargs['params'].update(model_config['params'])
+
             model_config.update(kwargs)
 
         return ModelConfig(**model_config)

mindmeld/components/entity_recognizer.py

@@ -17,7 +17,7 @@
 import logging
 import pickle
 
-import joblib
+from sklearn.externals import joblib
 
 from ._config import get_classifier_config
 from .classifier import Classifier, ClassifierConfig, ClassifierLoadError

mindmeld/components/role_classifier.py

@@ -17,7 +17,7 @@
 import logging
 import pickle
 
-import joblib
+from sklearn.externals import joblib
 
 from ._config import get_classifier_config
 from .classifier import Classifier, ClassifierConfig, ClassifierLoadError

mindmeld/models/model.py

@@ -23,7 +23,7 @@
 from inspect import signature
 from typing import Union, Type, Dict, Any, Tuple, List, Pattern, Set
 
-import joblib
+from sklearn.externals import joblib
 from sklearn.model_selection import (
     GridSearchCV,
     GroupKFold,
@@ -486,7 +486,7 @@ def _fit_cv(self, examples, labels, groups=None, selection_settings=None, fixed_
             n_jobs=n_jobs,
             return_train_score=False,
         )
-        model = grid_cv.fit(examples, y=labels, groups=groups)
+        model = grid_cv.fit(examples, labels, groups)
 
         for idx, params in enumerate(model.cv_results_["params"]):
             logger.debug("Candidate parameters: %s", params)

mindmeld/models/tagger_models.py

@@ -16,7 +16,7 @@
 import os
 import random
 
-import joblib
+from sklearn.externals import joblib
 
 from .evaluation import EntityModelEvaluation, EvaluatedExample
 from .helpers import (

mindmeld/models/taggers/lstm.py

@@ -15,9 +15,9 @@
 import os
 import re
 
-import joblib
 import numpy as np
 import tensorflow as tf
+from sklearn.externals import joblib
 from sklearn.preprocessing import LabelBinarizer
 
 from .embeddings import CharacterSequenceEmbedding, WordSequenceEmbedding

mindmeld/models/taggers/memm.py

@@ -150,7 +150,7 @@ def _predict_proba_example(self, example, config, resources):
             X, _ = self._preprocess_data([features])
             prediction = self._clf.predict_proba(X)[0]
             predicted_tag = np.argmax(prediction)
-            prev_tag = self.class_encoder.inverse_transform([predicted_tag])[0]
+            prev_tag = self.class_encoder.inverse_transform(predicted_tag)
             seq_log_probs.append([prev_tag, prediction[predicted_tag]])
         return seq_log_probs
 
@@ -175,7 +175,7 @@ def _predict_proba_distribution_example(self, example, config, resources):
             predictions.append(list(prediction))
             tag_maps.append(
                 [
-                    self.class_encoder.inverse_transform([i])[0] for i in range(len(prediction))
+                    self.class_encoder.inverse_transform(i) for i in range(len(prediction))
                 ]
             )
         return [tag_maps, predictions]
@@ -190,7 +190,7 @@ def _get_feature_selector(selector_type):
                 given the full feature matrix, X and the class labels, y.
         """
         selector = {
-            "l1": SelectFromModel(LogisticRegression(penalty="l1", C=1, solver="liblinear")),
+            "l1": SelectFromModel(LogisticRegression(penalty="l1", C=1)),
             "f": SelectPercentile(),
         }.get(selector_type)
         return selector

mindmeld/models/taggers/pytorch_crf.py

@@ -196,7 +196,7 @@ def get_tensor_data(self, feat_dicts, labels=None, fit=False):
             if isinstance(self.feat_extractor, DictVectorizer):
                 flattened_feat_dicts = list(chain.from_iterable(feat_dicts))
                 self.feat_extractor.fit(flattened_feat_dicts)
-                self.num_feats = len(self.feat_extractor.get_feature_names_out())
+                self.num_feats = len(self.feat_extractor.get_feature_names())
             if labels is not None:
                 flattened_labels = list(chain.from_iterable(labels))
                 self.label_encoder.fit(flattened_labels)
@@ -294,16 +294,8 @@ def save_best_weights_path(self, path):
         else:
             raise MindMeldError("CRF weights not saved. Please re-train model from scratch.")
 
-    def validate_params(self, kwargs):
+    def validate_params(self):
         """Validate the argument values saved into the CRF model. """
-        for key in kwargs:
-            msg = (
-                "Unexpected param `{param}`, dropping it from model config.".format(
-                    param=key
-                )
-            )
-            logger.warning(msg)
-
         if self.optimizer not in ["sgd", "adam"]:
             raise MindMeldError(
                 f"Optimizer type {self.optimizer_type} not supported. Supported options are ['sgd', 'adam']")
@@ -439,7 +431,7 @@ def compute_marginal_probabilities(self, inputs, mask):
     # pylint: disable=too-many-arguments
     def set_params(self, feat_type="hash", feat_num=50000, stratify_train_val_split=True, drop_input=0.2, batch_size=8,
                    number_of_epochs=100, patience=3, dev_split_ratio=0.2, optimizer="sgd",
-                   random_state=None, **kwargs):
+                   random_state=None):
         """Set the parameters for the PyTorch CRF model and also validates the parameters.
 
         Args:
@@ -467,7 +459,7 @@ def set_params(self, feat_type="hash", feat_num=50000, stratify_train_val_split=
         self.optimizer = optimizer  # ["sgd", "adam"]
         self.random_state = random_state or randint(1, 10000001)
 
-        self.validate_params(kwargs)
+        self.validate_params()
 
         logger.debug("Random state for torch-crf is %s", self.random_state)
         if self.feat_type == "dict":

mindmeld/models/text_models.py

@@ -20,9 +20,9 @@
 import os
 import random
 
-import joblib
 import numpy as np
 from sklearn.ensemble import RandomForestClassifier
+from sklearn.externals import joblib
 from sklearn.feature_extraction import DictVectorizer
 from sklearn.feature_selection import SelectFromModel, SelectPercentile
 from sklearn.linear_model import LogisticRegression
@@ -371,7 +371,7 @@ def _get_feature_selector(self):
         else:
             selector_type = self.config.model_settings.get("feature_selector")
         selector = {
-            "l1": SelectFromModel(LogisticRegression(penalty="l1", C=1, solver="liblinear")),
+            "l1": SelectFromModel(LogisticRegression(penalty="l1", C=1)),
             "f": SelectPercentile(),
         }.get(selector_type)
         return selector

setup.py

@@ -28,7 +28,7 @@
     "pytz",  # uses calendar versioning
     "scipy>=0.13.3,<2.0",
     'scikit-learn>=0.18.1,<0.20; python_version < "3.7"',
-    'scikit-learn~=1.0.2; python_version >= "3.7"',
+    'scikit-learn>=0.19.2,<0.20; python_version >= "3.7"',
     "requests>=2.20.1,<3.0",
     "tqdm~=4.15",
     'python-crfsuite~=0.9; python_version < "3.7"',

tests/models/test_tagging.py

@@ -271,8 +271,7 @@ def test_get_boundary_counts_sequential(
 
 @pytest.mark.parametrize(
     "model_type,params",
-    [("memm", {"penalty": "l2", "C": 10000}),
-     ("torch-crf", {"feat_type": "dict"}),
+    [("memm", {"penalty": "l2", "C": 10000}), ("crf", {"c1": 0.01, "c2": 0.01}), ("torch-crf", {"feat_type": "dict"}),
      ("torch-crf", {"feat_type": "hash"})],
 )
 def test_view_extracted_features(kwik_e_mart_nlp, model_type, params):
@@ -312,6 +311,7 @@ def test_view_extracted_features(kwik_e_mart_nlp, model_type, params):
     "query,model_type,params",
     [
         ("Main st store hours", "memm", {"penalty": "l2", "C": 10000}),
+        ("Main st store hours", "crf", {"c1": 0.01, "c2": 0.01}),
         ("Main st store hours", "torch-crf", {"feat_type": "dict"}),
         ("Main st store hours", "torch-crf", {"feat_type": "hash"})
     ],