added textual entailment score #3
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jjmachan
reviewed
May 12, 2023
belar/metrics/factual.py
Outdated
| self.id2label = model_config["id2label"] | ||
|
|
||
| def name(self): | ||
| return "Entailment-Score" |
There was a problem hiding this comment.
Suggested change
| return "Entailment-Score" | |
| return "entailment" |
belar/metrics/factual.py
Outdated
Comment on lines
58
to
59
| ground_truth: t.Union[str, t.List[str]], | ||
| generated_text: t.Union[str, t.List[str]], |
There was a problem hiding this comment.
should we use the new version?
str | list[str]
whichever we choose, we have to be consistent
Co-authored-by: Jithin James <jamesjithin97@gmail.com>
jjmachan
approved these changes
May 12, 2023
2 tasks
jjmachan
added a commit
that referenced
this pull request
Oct 17, 2025
## Issue Link / Problem Description
This PR introduces a comprehensive, reusable testing infrastructure to
streamline the migration of legacy metrics to the modern collections
architecture.
**Problem**: Migrating metrics requires consistent validation that new
implementations match legacy behavior, but this process was ad-hoc and
time-consuming without standardized tooling.
**Solution**: A complete testing framework with:
- Step-by-step migration guide
- Configuration-driven validation notebook
- Shared test utilities and fixtures
- Dataset loading with safe fallbacks
## Changes Made
### 📋 Migration Documentation
- **`tests/e2e/metrics_migration/plan-for-metrics-migration.md`**
- Complete migration workflow (pre-migration → implementation → testing
→ finalization)
- Code templates for prompts, output models, and metric classes
- Validation criteria for different metric types (LLM-based,
embeddings-based, deterministic)
### 📓 Testing Notebook
- **`tests/e2e/metrics_migration/metric_score_diff.ipynb`** +
**`tests/notebooks/metric_score_diff.ipynb`**
- General-purpose, configuration-driven (edit 1 cell to test any metric)
- Concurrent dataset comparison (Amnesty QA + FIQA)
- Statistical analysis with 7-plot visualizations
- Automated validation criteria checking
### 🧪 Test Infrastructure
- **`tests/e2e/metrics_migration/base_migration_test.py`**
- `BaseMigrationTest` class with reusable test methods
- `run_e2e_compatibility_test()` - Compare legacy vs modern with
tolerance
- `run_metric_specific_test()` - Custom behavior validation
- **`tests/e2e/metrics_migration/conftest.py`**
- Shared pytest fixtures: `legacy_llm`, `modern_llm`,
`legacy_embeddings`, `modern_embeddings`
- Automatic API key validation and graceful skipping
- **`tests/e2e/metrics_migration/test_utils.py`**
- Helper utilities for migration tests
### 🏭 Component Factories
- **`tests/utils/llm_setup.py`**
- `create_legacy_llm()` / `create_modern_llm()` - LLM initialization for
both architectures
- `create_legacy_embeddings()` / `create_modern_embeddings()` -
Embeddings initialization
- `check_api_key()` - Validation utility
### ⚡ Optimized Comparison Engine
- **`tests/utils/metric_comparison.py`**
- `compare_metrics()` - Concurrent execution with configurable
parallelism
- Support for both parallel (independent) and sequential (dependent)
metric execution
- `ComparisonResult` dataclass with statistical analysis and pandas
export
### 📊 Dataset Utilities
- **`tests/e2e/test_dataset_utils.py`**
- `load_amnesty_dataset_safe()` - Amnesty QA dataset with local fallback
- `load_fiqa_dataset_safe()` - FIQA dataset with local fallback
- Embedded sample data for offline/CI testing
### ⚙️ Configuration
- **`.gitignore`** - Added `plan/` directory for migration planning docs
- **`CLAUDE.md`** - Updated with migration workflow guidance
## Testing
### Validation Status
- [x] Automated tests: Infrastructure provides framework for all future
migrations
- [x] Manual validation: Tested with Context Recall migration
### Test Results (Context Recall Migration)
| Dataset | Samples | Mean |Diff| | Within Tolerance | Status |
|---------|---------|---------------|------------------|--------|
| Amnesty QA | 20 | 0.0708 | 90% (18/20) | ✅ |
| FIQA | 30 | 0.0667 | 93.3% (28/30) | ✅ |
**Validation Criteria Met:**
- ✅ Mean |diff| < 0.15 (stricter than per-case tolerance)
- ✅ >90% within 0.2 tolerance for LLM-based metrics
- ✅ No systematic bias (mean diff < 0.05)
- ✅ Domain generalization confirmed across datasets
## Impact & Benefits
**Before this PR:**
- Ad-hoc testing with no standardized approach
- Difficult to validate score consistency
- Time-consuming manual comparison
- No reusable infrastructure
**After this PR:**
1. **Faster migrations** - Standardized workflow reduces implementation
time
2. **Higher quality** - Dataset-based validation catches issues early
3. **Consistency** - All metrics follow same validation process
4. **Reusability** - Shared utilities eliminate boilerplate
5. **Documentation** - Clear guide enables team-wide contribution
## Architecture
```
tests/
├── e2e/metrics_migration/
│ ├── plan-for-metrics-migration.md # Migration guide
│ ├── metric_score_diff.ipynb # Testing notebook
│ ├── base_migration_test.py # Base test class
│ ├── conftest.py # Shared fixtures
│ └── test_utils.py # Test helpers
├── e2e/test_dataset_utils.py # Dataset loading
└── utils/
├── llm_setup.py # Component factories
└── metric_comparison.py # Comparison engine
```
## Next Steps
This infrastructure enables the following PRs:
1. **PR #2** (depends on this): Context Recall migration
2. **PR #3** (depends on PR #2): Context Precision migration
3. **Future PRs**: Remaining metric migrations using this framework
---
**Note**: This PR contains **only infrastructure and documentation** -
no actual metric implementations. Metric migrations follow in subsequent
PRs.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add entailment score using NLI.