File size: 10,914 Bytes
1d10b0a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Dataset loader for RAG Bench datasets."""
import os
from typing import List, Dict, Optional
from datasets import load_dataset
import pandas as pd
from tqdm import tqdm


class RAGBenchLoader:
    """Load and manage RAG Bench datasets."""
    
    SUPPORTED_DATASETS = [
        'covidqa', 
        'cuad', 
        'delucionqa', 
        'emanual', 
        'expertqa', 
        'finqa', 
        'hagrid', 
        'hotpotqa', 
        'msmarco', 
        'pubmedqa', 
        'tatqa', 
        'techqa'
    ]
    
    def __init__(self, cache_dir: str = "./data_cache"):
        """Initialize the dataset loader.
        
        Args:
            cache_dir: Directory to cache downloaded datasets
        """
        self.cache_dir = cache_dir
        os.makedirs(cache_dir, exist_ok=True)
    
    def load_dataset(self, dataset_name: str, split: str = "test", 
                     max_samples: Optional[int] = None) -> List[Dict]:
        """Load a RAG Bench dataset from rungalileo/ragbench.
        
        Args:
            dataset_name: Name of the dataset to load
            split: Dataset split (train/validation/test)
            max_samples: Maximum number of samples to load
            
        Returns:
            List of dictionaries containing dataset samples
        """
        if dataset_name not in self.SUPPORTED_DATASETS:
            raise ValueError(f"Unsupported dataset: {dataset_name}. "
                           f"Supported: {self.SUPPORTED_DATASETS}")
        
        print(f"Loading {dataset_name} dataset ({split} split) from rungalileo/ragbench...")
        
        try:
            # Load from rungalileo/ragbench
            dataset = load_dataset("rungalileo/ragbench", dataset_name, split=split, 
                                   cache_dir=self.cache_dir)
            
            processed_data = []
            samples = dataset if max_samples is None else dataset.select(range(min(max_samples, len(dataset))))
            
            # Process the dataset
            for item in tqdm(samples, desc=f"Processing {dataset_name}"):
                processed_data.append(self._process_ragbench_item(item, dataset_name))
            
            print(f"Loaded {len(processed_data)} samples from {dataset_name}")
            return processed_data
            
        except Exception as e:
            print(f"Error loading {dataset_name}: {str(e)}")
            print("Falling back to sample data for testing...")
            return self._create_sample_data(dataset_name, max_samples or 10)
    
    def _process_ragbench_item(self, item: Dict, dataset_name: str) -> Dict:
        """Process a single RAGBench dataset item into standardized format.
        
        Args:
            item: Raw dataset item
            dataset_name: Name of the dataset
            
        Returns:
            Processed item dictionary
        """
        # RAGBench datasets typically have: question, documents, answer, and retrieved_contexts
        processed = {
            "question": item.get("question", ""),
            "answer": item.get("answer", ""),
            "context": "",  # For embedding and retrieval
            "documents": [],  # Store original documents list
            "dataset": dataset_name,
            "ground_truth_scores": {}  # NEW: Extract ground truth evaluation scores
        }
        
        # Extract documents - RAGBench uses 'documents' as primary source for embeddings
        # Priority: documents > retrieved_contexts > context
        if "documents" in item:
            if isinstance(item["documents"], list):
                processed["documents"] = [str(doc) for doc in item["documents"]]
                processed["context"] = " ".join(processed["documents"])
            else:
                processed["documents"] = [str(item["documents"])]
                processed["context"] = str(item["documents"])
        elif "retrieved_contexts" in item:
            if isinstance(item["retrieved_contexts"], list):
                processed["documents"] = [str(ctx) for ctx in item["retrieved_contexts"]]
                processed["context"] = " ".join(processed["documents"])
            else:
                processed["documents"] = [str(item["retrieved_contexts"])]
                processed["context"] = str(item["retrieved_contexts"])
        elif "context" in item:
            if isinstance(item["context"], list):
                processed["documents"] = [str(ctx) for ctx in item["context"]]
                processed["context"] = " ".join(processed["documents"])
            else:
                processed["documents"] = [str(item["context"])]
                processed["context"] = str(item["context"])
        
        # Extract ground truth evaluation scores from RAGBench dataset
        # These are pre-computed metrics from the RAGBench paper
        ground_truth_scores = {}
        
        # Extract metric scores - try multiple possible field names
        # RAGBench paper uses these metric names (with various possible field formats)
        score_fields = [
            # (possible_field_names, canonical_metric_name)
            (["relevance_score", "context_relevance", "relevance", "R"], "context_relevance"),
            (["utilization_score", "context_utilization", "utilization", "T"], "context_utilization"),
            (["completeness_score", "completeness", "C"], "completeness"),
            (["adherence_score", "adherence", "A", "overall_supported"], "adherence"),
        ]
        
        for field_names, metric_name in score_fields:
            for field_name in field_names:
                if field_name in item:
                    try:
                        # Handle string/numeric conversion
                        score_value = item[field_name]
                        if isinstance(score_value, bool):
                            # Boolean adherence: True=1.0, False=0.0
                            score_value = 1.0 if score_value else 0.0
                        elif isinstance(score_value, str):
                            # Try to convert string to float
                            if score_value.lower() in ['true', 'yes']:
                                score_value = 1.0
                            elif score_value.lower() in ['false', 'no']:
                                score_value = 0.0
                            else:
                                score_value = float(score_value)
                        ground_truth_scores[metric_name] = float(score_value)
                        break  # Found this metric, move to next
                    except (ValueError, TypeError):
                        continue  # Try next field name
        
        # Store ground truth scores if any were found
        if ground_truth_scores:
            processed["ground_truth_scores"] = ground_truth_scores
        
        # Store additional metadata if available
        if "metadata" in item:
            processed["metadata"] = item["metadata"]
        
        return processed
    
    def load_all_datasets(self, split: str = "test", max_samples: Optional[int] = None) -> Dict[str, List[Dict]]:
        """Load all RAGBench datasets.
        
        Args:
            split: Dataset split to load
            max_samples: Maximum samples per dataset
            
        Returns:
            Dictionary mapping dataset names to their data
        """
        all_data = {}
        for dataset_name in self.SUPPORTED_DATASETS:
            print(f"\n{'='*50}")
            print(f"Loading {dataset_name}...")
            print(f"{'='*50}")
            try:
                all_data[dataset_name] = self.load_dataset(dataset_name, split, max_samples)
            except Exception as e:
                print(f"Failed to load {dataset_name}: {str(e)}")
                all_data[dataset_name] = []
        
        return all_data
    
    def _create_sample_data(self, dataset_name: str, num_samples: int) -> List[Dict]:
        """Create sample data for testing when actual dataset is unavailable."""
        sample_data = []
        for i in range(num_samples):
            # Create multiple sample documents per question
            sample_docs = [
                f"Document 1: This is the first sample document {i+1} for {dataset_name} dataset. "
                f"It contains relevant information to answer the question.",
                f"Document 2: This is the second sample document {i+1} providing additional context. "
                f"It includes more details about the topic.",
                f"Document 3: This is the third sample document {i+1} with supplementary information."
            ]
            
            sample_data.append({
                "question": f"Sample question {i+1} for {dataset_name}?",
                "answer": f"Sample answer {i+1}",
                "documents": sample_docs,
                "context": " ".join(sample_docs),  # Combined for backward compatibility
                "dataset": dataset_name
            })
        return sample_data
    
    def get_test_data(self, dataset_name: str, num_samples: int = 100) -> List[Dict]:
        """Get test data for TRACE evaluation.
        
        Args:
            dataset_name: Name of the dataset
            num_samples: Number of test samples
            
        Returns:
            List of test samples
        """
        return self.load_dataset(dataset_name, split="test", max_samples=num_samples)
    
    def get_test_data_size(self, dataset_name: str) -> int:
        """Get the total number of test samples available in a dataset.
        
        Args:
            dataset_name: Name of the dataset
            
        Returns:
            Total number of test samples available
        """
        try:
            from datasets import load_dataset_builder
            
            # Load dataset builder to get dataset info
            builder = load_dataset_builder("rungalileo/ragbench", dataset_name)
            
            # Try to get test split size
            if hasattr(builder.info, 'splits') and builder.info.splits:
                if 'test' in builder.info.splits:
                    return builder.info.splits['test'].num_examples
                elif 'validation' in builder.info.splits:
                    return builder.info.splits['validation'].num_examples
                else:
                    # Get first available split
                    first_split = list(builder.info.splits.keys())[0]
                    return builder.info.splits[first_split].num_examples
            
            # Fallback: load full test dataset to count
            ds = load_dataset("rungalileo/ragbench", dataset_name, split="test", trust_remote_code=True)
            return len(ds)
        except Exception as e:
            print(f"Error getting test data size: {e}")
            # Return a reasonable default
            return 100