File size: 14,683 Bytes
714cf46
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
import numpy as np
import pandas as pd
from sklearn.model_selection import RandomizedSearchCV
from typing import Dict, Any, Tuple, Optional

try:
    from metrics import (
        get_regression_scorer, get_classification_scorer,
        classification_scorer, regression_scorer,
        compute_single_label_classification_metrics,
        compute_regression_metrics,
    )
    from utils import print_message
    from seed_utils import get_global_seed
except ImportError:
    from ..metrics import (
        get_regression_scorer, get_classification_scorer,
        classification_scorer, regression_scorer,
        compute_single_label_classification_metrics,
        compute_regression_metrics,
    )
    from ..utils import print_message
    from ..seed_utils import get_global_seed

from transformers import EvalPrediction
from .lazy_predict import (
    LazyRegressor,
    LazyClassifier,
    CLASSIFIER_DICT,
    REGRESSOR_DICT,
    ALL_MODEL_DICT
)
from .scikit_hypers import HYPERPARAMETER_DISTRIBUTIONS


class ScikitArguments:
    """
    Combined arguments class for scikit-learn model training and tuning.
    """
    def __init__(
        self,
        # Tuning arguments
        n_iter: int = 100,
        cv: int = 3,
        random_state: Optional[int] = None,
        # Specific model arguments (optional)
        model_name: Optional[str] = None,
        scikit_model_name: Optional[str] = None,  # CLI arg name
        scikit_model_args: Optional[str] = None,  # CLI arg - JSON string
        model_args: Optional[Dict[str, Any]] = None,
        production_model: bool = False,
        **kwargs,
    ):
        import json
        # Tuning arguments
        self.n_iter = n_iter
        self.cv = cv
        self.random_state = random_state or get_global_seed()
        
        # Specific model arguments - scikit_model_name takes precedence (CLI arg)
        self.model_name = scikit_model_name or model_name
        
        # Parse scikit_model_args JSON string if provided (CLI), otherwise use model_args dict
        if scikit_model_args is not None:
            try:
                self.model_args = json.loads(scikit_model_args)
                print_message(f"Using pre-specified hyperparameters (skipping tuning): {self.model_args}")
            except json.JSONDecodeError as e:
                raise ValueError(f"Failed to parse --scikit_model_args JSON: {e}")
        else:
            self.model_args = model_args if model_args is not None else {}
        
        self.production_model = production_model


class ModelResults:
    def __init__(
        self,
        initial_scores: Optional[pd.DataFrame],
        best_model_name: str,
        best_params: Optional[Dict[str, Any]],
        final_scores: Dict[str, float],
        best_model: Any
    ):
        self.initial_scores = initial_scores
        self.best_model_name = best_model_name
        self.best_params = best_params
        self.final_scores = final_scores
        self.best_model = best_model

    def __str__(self) -> str:
        return (
            f"Best Model: {self.best_model_name}\n"
            f"Best Parameters: {self.best_params}\n"
            f"Final Scores: {self.final_scores}"
        )


class ScikitProbe:
    """
    A class for finding and tuning the best scikit-learn models for a given dataset.
    """
    def __init__(self, args: ScikitArguments):
        self.args = args
        self.n_jobs = 1
    
    def _tune_hyperparameters(
        self,
        model_class: Any,
        model_name: str,
        X_train: np.ndarray,
        y_train: np.ndarray,
        custom_scorer: Any,
    ) -> Tuple[Any, Dict[str, Any]]:
        """
        Perform hyperparameter tuning using RandomizedSearchCV.
        """
        param_distributions = HYPERPARAMETER_DISTRIBUTIONS.get(model_name, {})
        if not param_distributions:
            print_message(f"No hyperparameter distributions defined for {model_name}, using defaults")
            return model_class(), {}

        print_message(f"Running RandomizedSearchCV with {self.args.n_iter} iterations, {self.args.cv}-fold CV...")
        print_message(f"Hyperparameter search space: {list(param_distributions.keys())}")
        
        random_search = RandomizedSearchCV(
            model_class(),
            param_distributions=param_distributions,
            n_iter=self.args.n_iter,
            scoring=custom_scorer,
            cv=self.args.cv,
            random_state=self.args.random_state,
            n_jobs=self.n_jobs,
            verbose=2  # Show progress for each fit
        )
        
        random_search.fit(X_train, y_train)
        print_message(f"Best CV score: {random_search.best_score_:.4f}")
        return random_search.best_estimator_, random_search.best_params_

    def find_best_regressor(
        self,
        X_train: np.ndarray,
        y_train: np.ndarray,
        X_test: np.ndarray,
        y_test: np.ndarray,
    ) -> ModelResults:
        """
        Find the best regression model through lazy prediction and hyperparameter tuning.
        
        Args:
            X_train: Training features
            y_train: Training targets
            X_test: Test features
            y_test: Test targets
            
        Returns:
            ModelResults object containing all results and the best model
        """
        # Initial lazy prediction
        print_message(f"Initial lazy prediction started")
        regressor = LazyRegressor(
            verbose=0,
            ignore_warnings=False,
            custom_metric=regression_scorer()
        )
        initial_scores = regressor.fit(X_train, X_test, y_train, y_test)
        if isinstance(initial_scores, Tuple):
            initial_scores = initial_scores[0]
        
        # Get best model name and class
        best_model_name = initial_scores.index[0]
        # Models are now stored directly (not as Pipeline) after optimization
        best_model_class = regressor.models[best_model_name].__class__
        print_message(f"Best model name: {best_model_name}")
        print_message(f"Best model class: {best_model_class}")
        print_message(f"Initial scores: \n{initial_scores}")

        print_message(f"Tuning hyperparameters")
        # Tune hyperparameters
        scorer = get_regression_scorer()
        best_model, best_params = self._tune_hyperparameters(
            best_model_class,
            best_model_name,
            X_train,
            y_train,
            scorer,
        )
        
        # Get final scores with tuned model
        best_model.fit(X_train, y_train)
        final_scores = self._calculate_metrics(best_model, X_test, y_test, best_model_name)
        print_message(f"Final scores: {final_scores}")
        print_message(f"Best params: \n{best_params}")

        return ModelResults(
            initial_scores=initial_scores,
            best_model_name=best_model_name,
            best_params=best_params,
            final_scores=final_scores,
            best_model=best_model
        )

    def find_best_classifier(
        self,
        X_train: np.ndarray,
        y_train: np.ndarray,
        X_test: np.ndarray,
        y_test: np.ndarray,
    ) -> ModelResults:
        """
        Find the best classification model through lazy prediction and hyperparameter tuning.
        
        Args:
            X_train: Training features
            y_train: Training targets
            X_test: Test features
            y_test: Test targets
            
        Returns:
            ModelResults object containing all results and the best model
        """
        # Initial lazy prediction
        print_message(f"Initial lazy prediction started")
        classifier = LazyClassifier(
            verbose=0,
            ignore_warnings=False,
            custom_metric=classification_scorer()
        )
        initial_scores = classifier.fit(X_train, X_test, y_train, y_test)
        if isinstance(initial_scores, Tuple):
            initial_scores = initial_scores[0]

        # Get best model name and class
        best_model_name = initial_scores.index[0]
        # Models are now stored directly (not as Pipeline) after optimization
        best_model_class = classifier.models[best_model_name].__class__
        print_message(f"Best model name: {best_model_name}")
        print_message(f"Best model class: {best_model_class}")
        print_message(f"Initial scores: \n{initial_scores}")

        print_message(f"Tuning hyperparameters")
        # Tune hyperparameters
        scorer = get_classification_scorer()
        best_model, best_params = self._tune_hyperparameters(
            best_model_class,
            best_model_name,
            X_train,
            y_train,
            scorer,
        )
        
        # Get final scores with tuned model
        best_model.fit(X_train, y_train)
        final_scores = self._calculate_metrics(best_model, X_test, y_test, best_model_name)
        print_message(f"Final scores: {final_scores}")
        print_message(f"Best params: \n{best_params}")

        return ModelResults(
            initial_scores=initial_scores,
            best_model_name=best_model_name,
            best_params=best_params,
            final_scores=final_scores,
            best_model=best_model
        )

    def _calculate_metrics(
        self,
        model: Any,
        X: np.ndarray,
        y: np.ndarray,
        model_name: str,
    ) -> Dict[str, float]:
        """
        Delegate to the shared metric functions in metrics.py via EvalPrediction,
        keeping a single source of truth for metric calculation across the codebase.
        """
        if model_name in CLASSIFIER_DICT:
            if hasattr(model, 'predict_proba'):
                predictions = model.predict_proba(X)
            else:
                # Fall back to one-hot hard predictions for models without predict_proba
                y_pred = model.predict(X)
                n_classes = len(np.unique(y))
                predictions = np.eye(n_classes)[y_pred.astype(int)]
            p = EvalPrediction(predictions=predictions, label_ids=y)
            return compute_single_label_classification_metrics(p)

        elif model_name in REGRESSOR_DICT:
            y_pred = model.predict(X)
            p = EvalPrediction(predictions=y_pred, label_ids=y)
            return compute_regression_metrics(p)

        return {}

    def run_specific_model(
        self,
        X_train: np.ndarray,
        y_train: np.ndarray,
        X_valid: np.ndarray,
        y_valid: np.ndarray,
        X_test: np.ndarray,
        y_test: np.ndarray,
        model_results: Optional[ModelResults] = None,
    ) -> ModelResults:
        """
        Run a specific model with given arguments or based on a previous ModelResults.
        
        Args:
            X_train: Training features
            y_train: Training targets
            X_valid: Validation features
            y_valid: Validation targets
            X_test: Test features
            y_test: Test targets
            model_results: Optional ModelResults from find_best_classifier or find_best_regressor
                          If provided, will use the best model type and parameters from it
            
        Returns:
            ModelResults object containing results and the model
        """
        print_message("Running specific model")
        if self.args.production_model:
            print_message(f"Running in production mode, train and validation are combined")
            X_train = np.concatenate([X_train, X_valid])
            y_train = np.concatenate([y_train, y_valid])

        # If model_results is provided, use its best model type and parameters
        if model_results is not None:
            model_name = model_results.best_model_name
            model_params = model_results.best_params if model_results.best_params is not None else {}
            
            # Get the model class
            model_class = ALL_MODEL_DICT[model_name]
            
            # Create and train the model with the best parameters
            cls = model_class(**model_params)
            print_message(f"Training model {cls}")
            cls.fit(X_train, y_train)
            print_message(f"Model trained")
            
            final_scores = self._calculate_metrics(cls, X_test, y_test, model_name)
            print_message(f"Final scores: {final_scores}")

            return ModelResults(
                initial_scores=None,
                best_model_name=model_name,
                best_params=model_params,
                final_scores=final_scores,
                best_model=cls
            )
        
        # Original functionality when no model_results is provided
        elif self.args.model_name is not None:
            model_name = self.args.model_name
            if model_name in CLASSIFIER_DICT:
                scorer = get_classification_scorer()
            elif model_name in REGRESSOR_DICT:
                scorer = get_regression_scorer()
            else:
                raise ValueError(f"Model {model_name} not supported")

            model_class = ALL_MODEL_DICT[model_name]
            
            # Skip tuning if model_args is already provided
            if self.args.model_args:
                print_message(f"Skipping hyperparameter tuning - using provided args: {self.args.model_args}")
                best_model = model_class(**self.args.model_args)
                best_params = self.args.model_args
            else:
                # Run hyperparameter tuning
                print_message(f"Tuning hyperparameters for {model_name}")
                best_model, best_params = self._tune_hyperparameters(
                    model_class,
                    model_name,
                    X_train,
                    y_train,
                    scorer
                )
                print_message(f"Best parameters: {best_params}")
            
            # Train final model with best parameters
            print_message(f"Training final model with best parameters")
            best_model.fit(X_train, y_train)
            
            final_scores = self._calculate_metrics(best_model, X_test, y_test, model_name)
            print_message(f"Final scores: {final_scores}")
            
            return ModelResults(
                initial_scores=None,
                best_model_name=model_name,
                best_params=best_params,
                final_scores=final_scores,
                best_model=best_model
            )
        else:
            raise ValueError("Either model_name must be specified in args or model_results must be provided")