File size: 14,873 Bytes
e2a0b30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
399
400
401
402
403
404
import ast

from transformers import AutoModelForMaskedLM, AutoModelForCausalLM, AutoTokenizer
from transformers import PreTrainedModel, PreTrainedTokenizerBase
from typing import Tuple, NamedTuple, List, Any
import torch.nn.functional as F
from tqdm import tqdm
import pandas as pd
import itertools
import logging
import torch
import math

from grewtse.utils.validation import load_and_validate_mp_dataset
from grewtse.evaluators.metrics import (
    compute_normalised_surprisal_difference,
    compute_average_surprisal_difference,
    compute_entropy,
    compute_surprisal,
    compute_mean,
    calculate_all_metrics
)

EVAL_TEMPLATE = {
    "sentence_id": None,
    "match_id": None,
    "original_text": None,
    "prompt_text": None,
    "form_grammatical": None,
    "p_grammatical": None,
    "I_grammatical": None,
    "form_ungrammatical": None,
    "p_ungrammatical": None,
    "I_ungrammatical": None,
    "entropy": None,
    "entropy_norm": None,
}


class TooManyMasksException(Exception):
    def __init__(self, message: str):
        self.message = message
        super().__init__(f"TMM Exception: {message}")


class Prediction(NamedTuple):
    token: str
    prob: float
    surprisal: float


class GrewTSEvaluator:
    """
    An evaluation class designed specifically for rapid syntactic evaluation of models available on the Hugging Face platform.
    """

    def __init__(self):
        self.evaluator = Evaluator()
        self.evaluation_dataset = None

    def evaluate_model(
        self,
        mp_dataset: pd.DataFrame,
        model_repo: str,
        model_type: str,  # can be 'encoder' or 'decoder'
        entropy_topk: int = 100,
        row_limit: int = None,
    ) -> pd.DataFrame:
        """
        Generic evaluation function for encoder or decoder models.
        """

        # --- Prepare dataset ---
        mp_dataset_iter = mp_dataset.itertuples()
        if row_limit:
            mp_dataset_iter = itertools.islice(mp_dataset_iter, row_limit)
        n = len(mp_dataset) if not row_limit else row_limit

        # --- Load model & tokenizer ---
        is_encoder = model_type == "encoder"
        model, tokenizer = self.evaluator.setup_parameters(model_repo, is_encoder)
        results = []

        # --- Evaluate each row ---
        for row in tqdm(mp_dataset_iter, ncols=n):
            row_results = self._init_row_results(row)

            try:
                if is_encoder:
                    self._evaluate_encoder_row(row, row_results)
                else:
                    self._evaluate_decoder_row(row, row_results)

            except TooManyMasksException:
                logging.error(f"Too many masks in {row.sentence_id}")
                continue
            except Exception as e:
                raise RuntimeError(f"Model/tokeniser issue: {e}") from e

            # --- Entropy ---
            entropy, entropy_norm = self.evaluator.get_entropy(entropy_topk, True)
            row_results["entropy"] = entropy
            row_results["entropy_norm"] = entropy_norm

            results.append(row_results)

        results_df = pd.DataFrame(results, columns=EVAL_TEMPLATE.keys())
        self.evaluation_dataset = results_df
        return results_df

    def evaluate_from_minimal_pairs(
        self,
        mp_dataset_filepath: str,
        model_repo: str,
        model_type: str,
        entropy_topk: int = 100,
        row_limit: int = None,
    ) -> pd.DataFrame:
        mp_dataset = load_and_validate_mp_dataset(mp_dataset_filepath)
        self.mp_dataset = mp_dataset
        return self.evaluate_model(model_repo, model_type, entropy_topk, row_limit)

    # --- Helper functions ---
    def _init_row_results(self, row):
        row_results = EVAL_TEMPLATE.copy()
        row_results.update(row._asdict())
        return row_results

    def _evaluate_encoder_row(self, row, row_results):
        prob_gram, prob_ungram = self.evaluator.run_masked_prediction(
            row.masked_text,
            row.form_grammatical,
            row.form_ungrammatical,
        )

        row_results["p_grammatical"] = prob_gram
        row_results["p_ungrammatical"] = prob_ungram
        row_results["I_grammatical"] = compute_surprisal(prob_gram)
        row_results["I_ungrammatical"] = compute_surprisal(prob_ungram)

        if "ood_minimal_pairs" in row:
            ood_pairs_str = row.ood_pairs
            ood_pairs = ast.literal_eval(ood_pairs_str)
            all_ood_probs_gram = []
            all_ood_probs_ungram = []
            for pair in ood_pairs:
                ood_prob_gram, ood_prob_ungram = self.evaluator.run_masked_prediction(
                    row.masked_text, pair[0], pair[1]
                )
                all_ood_probs_gram.append(ood_prob_gram)
                all_ood_probs_ungram.append(ood_prob_ungram)

            avg_ood_prob_gram = compute_mean(all_ood_probs_gram)
            avg_ood_prob_ungram = compute_mean(all_ood_probs_ungram)

            row_results["ood_p_grammatical"] = avg_ood_prob_gram
            row_results["ood_p_ungrammatical"] = avg_ood_prob_ungram
            row_results["ood_I_grammatical"] = compute_surprisal(avg_ood_prob_gram)
            row_results["ood_I_ungrammatical"] = compute_surprisal(avg_ood_prob_ungram)

    def _evaluate_decoder_row(self, row, row_results):
        prob_gram, prob_ungram = self.evaluator.run_next_word_prediction(
            row.prompt_text, row.form_grammatical, row.form_ungrammatical
        )
        row_results["p_grammatical"] = prob_gram
        row_results["p_ungrammatical"] = prob_ungram
        row_results["I_grammatical"] = compute_surprisal(prob_gram)
        row_results["I_ungrammatical"] = compute_surprisal(prob_ungram)

        if "ood_minimal_pairs" in row:
            ood_pairs_str = row.ood_pairs
            ood_pairs = ast.literal_eval(ood_pairs_str)
            all_ood_probs_gram = []
            all_ood_probs_ungram = []
            for pair in ood_pairs:
                ood_prob_gram, ood_prob_ungram = (
                    self.evaluator.run_next_word_prediction(
                        row.masked_text, pair[0], pair[1]
                    )
                )
                all_ood_probs_gram.append(ood_prob_gram)
                all_ood_probs_ungram.append(ood_prob_ungram)

            avg_ood_prob_gram = compute_mean(all_ood_probs_gram)
            avg_ood_prob_ungram = compute_mean(all_ood_probs_ungram)

            row_results["ood_p_grammatical"] = avg_ood_prob_gram
            row_results["ood_p_ungrammatical"] = avg_ood_prob_ungram
            row_results["ood_I_grammatical"] = compute_surprisal(avg_ood_prob_gram)
            row_results["ood_I_ungrammatical"] = compute_surprisal(avg_ood_prob_ungram)

    def get_norm_avg_surprisal_difference(self) -> float:
        if not self.is_model_evaluated():
            raise KeyError("Please evaluate a model first.")
        return compute_normalised_surprisal_difference(
            self.evaluation_dataset["p_grammatical"],
            self.evaluation_dataset["p_ungrammatical"],
        )

    def get_avg_surprisal_difference(self, is_ood: bool = False) -> float:
        p_grammatical_col = "p_grammatical" if not is_ood else "ood_p_grammatical"
        p_ungrammatical_col = "p_ungrammatical" if not is_ood else "ood_p_ungrammatical"
        if not self.is_model_evaluated():
            raise KeyError("Please evaluate a model first.")
        return compute_average_surprisal_difference(
            self.evaluation_dataset[p_grammatical_col],
            self.evaluation_dataset[p_ungrammatical_col],
        )

    def get_all_metrics(self):
        if self.evaluation_dataset is not None:
            print(self.evaluation_dataset.columns)
            return calculate_all_metrics(self.evaluation_dataset)
        else:
            raise ValueError("Please evaluate a model first.")


class Evaluator:
    def __init__(self):
        self.tokeniser: PreTrainedTokenizerBase = None
        self.model: PreTrainedModel = None

        self.mask_token_index: int = -1
        self.mask_probs: torch.Tensor | None = None
        self.logits: torch.Tensor = None

    def setup_parameters(
        self, model_name: str, is_mlm: bool = True
    ) -> Tuple[PreTrainedTokenizerBase, PreTrainedModel]:
        if is_mlm:
            self.tokeniser = AutoTokenizer.from_pretrained(model_name)
            self.model = AutoModelForMaskedLM.from_pretrained(model_name)
        else:
            self.tokeniser = AutoTokenizer.from_pretrained(model_name)
            self.model = AutoModelForCausalLM.from_pretrained(model_name)

        # set to eval mode, disabling things like dropout
        self.model.eval()

        return self.model, self.tokeniser

    def run_masked_prediction(
        self, sentence: str, grammatical_word: str, ungrammatical_word: str
    ) -> Tuple[float, float]:
        if not self.model or not self.tokeniser:
            raise RuntimeError("Model and tokenizer must be loaded before prediction.")

        mask_token = self.tokeniser.mask_token
        sentence_masked = sentence.replace("[MASK]", mask_token)

        if sentence_masked.count(mask_token) != 1:
            raise TooManyMasksException("Only single-mask sentences are supported.")

        masked_ids = self.tokeniser.encode(sentence_masked, add_special_tokens=False)
        mask_index = masked_ids.index(self.tokeniser.mask_token_id)

        device = next(self.model.parameters()).device
        g_ids = self.tokeniser.encode(grammatical_word, add_special_tokens=False)
        u_ids = self.tokeniser.encode(ungrammatical_word, add_special_tokens=False)

        g_prob = self._compute_masked_joint_probability(
            masked_ids, mask_index, g_ids, device
        )
        u_prob = self._compute_masked_joint_probability(
            masked_ids, mask_index, u_ids, device
        )

        return g_prob, u_prob

    def _compute_masked_joint_probability(
        self, input_ids: List[int], mask_index: int, word_ids: List[int], device
    ) -> float:
        input_ids_tensor = torch.tensor([input_ids], device=device)
        log_prob = 0.0
        index = mask_index

        for i, tid in enumerate(word_ids):
            with torch.no_grad():
                logits = self.model(input_ids_tensor).logits

            probs = F.softmax(logits[:, index, :], dim=-1)
            token_prob = probs[0, tid].item()
            log_prob += math.log(token_prob + 1e-12)

            if i == 0:
                self.mask_probs = probs

            # Replace mask with predicted token
            input_ids_tensor[0, index] = tid

            # Insert new mask if more tokens remain
            if i < len(word_ids) - 1:
                input_ids_tensor = torch.cat(
                    [
                        input_ids_tensor[:, : index + 1],
                        torch.tensor([[self.tokeniser.mask_token_id]], device=device),
                        input_ids_tensor[:, index + 1 :],
                    ],
                    dim=1,
                )

                # debugging
                tokens_after_insertion = self.tokeniser.convert_ids_to_tokens(
                    input_ids_tensor[0].tolist()
                )

                index += 1

        return math.exp(log_prob)

    def run_next_word_prediction(
        self, context: str, grammatical_word: str, ungrammatical_word: str
    ) -> Tuple[float, float]:
        if not self.model or not self.tokeniser:
            raise RuntimeError("Model and tokenizer must be loaded before prediction.")

        context_ids = self.tokeniser.encode(context, add_special_tokens=False)
        device = next(self.model.parameters()).device

        g_ids = self.tokeniser.encode(grammatical_word, add_special_tokens=False)
        u_ids = self.tokeniser.encode(ungrammatical_word, add_special_tokens=False)

        g_prob = self._compute_next_word_joint_probability(context_ids, g_ids, device)
        u_prob = self._compute_next_word_joint_probability(context_ids, u_ids, device)

        return g_prob, u_prob

    def _compute_next_word_joint_probability(
        self, input_ids: List[int], word_ids: List[int], device
    ) -> float:
        input_ids_tensor = torch.tensor([input_ids], device=device)
        # debugging
        tokens_after_insertion = self.tokeniser.convert_ids_to_tokens(
            input_ids_tensor[0].tolist()
        )
        log_prob = 0.0

        for i, tid in enumerate(word_ids):
            with torch.no_grad():
                logits = self.model(input_ids_tensor).logits

            index = input_ids_tensor.shape[1] - 1  # last token position
            probs = F.softmax(logits[:, index, :], dim=-1)
            token_prob = probs[0, tid].item()
            log_prob += math.log(token_prob + 1e-12)

            if i == 0:
                self.mask_probs = probs

            # Append predicted token to context
            input_ids_tensor = torch.cat(
                [input_ids_tensor, torch.tensor([[tid]], device=device)], dim=1
            )

            # debugging
            tokens_after_insertion = self.tokeniser.convert_ids_to_tokens(
                input_ids_tensor[0].tolist()
            )

        return math.exp(log_prob)

    def get_entropy(self, k: int = 100, normalise: bool = False) -> float:
        """Compute entropy over the prediction distribution.

        Args:
            k: Number of top tokens to consider.
            normalise: Whether to normalise entropy.

        Returns:
            Entropy value.
        Raises:
            ValueError: If no probabilities are available.
        """
        if self.mask_probs is None:
            raise ValueError("No output probabilities available. Run evaluation first.")
        return compute_entropy(self.mask_probs, k, normalise)

    def _get_mask_index(self, inputs: Any) -> int:
        if "input_ids" not in inputs:
            raise ValueError("Missing 'input_ids' in inputs.")
        elif self.tokeniser.mask_token_id is None:
            raise ValueError("The tokeniser does not have a defined mask_token_id.")

        input_ids = inputs["input_ids"]
        mask_positions = torch.where(input_ids == self.tokeniser.mask_token_id)

        if len(mask_positions[0]) == 0:
            raise ValueError("No mask token found in input_ids.")
        elif len(mask_positions[0]) > 1:
            raise ValueError("Multiple mask tokens found; expected only one.")

        return (
            mask_positions[1].item()
            if len(mask_positions) > 1
            else mask_positions[0].item()
        )

    def _get_mask_probabilities(
        self, mask_token_index: int, logits: Any
    ) -> torch.Tensor:
        mask_logits = logits[0, mask_token_index, :]
        probs = F.softmax(mask_logits, dim=-1)  # shape: (vocab_size, )
        return probs