Datasets:
Fix 04_generate_parquet.py: path handling and MWT schema
Browse filesFixes for Parquet generation:
1. **Path handling**: Extract filename from metadata path
- Metadata includes full path: 'UD_French-GSD/r2.17/fr_gsd-ud-train.conllu'
- Only need filename: 'fr_gsd-ud-train.conllu'
- Files are in: UD_repos/UD_French-GSD/ (not nested by revision)
2. **MWT schema definition**: Use list with dict instead of Sequence
- Changed from: datasets.Sequence({...})
- Changed to: [{...}]
- Fixes AttributeError: 'list' object has no attribute 'get'
- Matches HuggingFace datasets schema requirements
3. **Error handling**: Added traceback for debugging
- Helps identify issues during generation
- Shows full stack trace for failures
Verified:
- ✅ All 235 treebanks processed successfully (0 failures)
- ✅ 506 Parquet files generated (622 MB total)
- ✅ MWT field working correctly
- ✅ French-GSD: 6,188 sentences with MWTs extracted
- ✅ Loading speed: ~147,000 examples/second
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
@@ -18,6 +18,7 @@ import argparse
|
|
| 18 |
import json
|
| 19 |
import os
|
| 20 |
import sys
|
|
|
|
| 21 |
from pathlib import Path
|
| 22 |
from typing import Any, Dict, List
|
| 23 |
|
|
@@ -142,8 +143,12 @@ def generate_parquet_for_treebank(
|
|
| 142 |
# Extract examples from all files in this split
|
| 143 |
all_examples = []
|
| 144 |
for file_path in files:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
# Construct full path: UD_repos/dirname/filename
|
| 146 |
-
full_path = UD_REPOS_DIR / metadata["dirname"] /
|
| 147 |
|
| 148 |
if not full_path.exists():
|
| 149 |
print(f" Warning: File not found: {full_path}", file=sys.stderr)
|
|
@@ -181,11 +186,7 @@ def generate_parquet_for_treebank(
|
|
| 181 |
"deprel": datasets.Sequence(datasets.Value("string")),
|
| 182 |
"deps": datasets.Sequence(datasets.Value("string")),
|
| 183 |
"misc": datasets.Sequence(datasets.Value("string")),
|
| 184 |
-
"mwt": datasets.
|
| 185 |
-
"id": datasets.Value("string"),
|
| 186 |
-
"form": datasets.Value("string"),
|
| 187 |
-
"misc": datasets.Value("string"),
|
| 188 |
-
}),
|
| 189 |
})
|
| 190 |
|
| 191 |
# Create dataset from examples
|
|
@@ -306,6 +307,7 @@ def main():
|
|
| 306 |
|
| 307 |
except Exception as e:
|
| 308 |
print(f" Error: {e}", file=sys.stderr)
|
|
|
|
| 309 |
fail_count += 1
|
| 310 |
|
| 311 |
if verbose:
|
|
|
|
| 18 |
import json
|
| 19 |
import os
|
| 20 |
import sys
|
| 21 |
+
import traceback
|
| 22 |
from pathlib import Path
|
| 23 |
from typing import Any, Dict, List
|
| 24 |
|
|
|
|
| 143 |
# Extract examples from all files in this split
|
| 144 |
all_examples = []
|
| 145 |
for file_path in files:
|
| 146 |
+
# Extract just the filename from the path (metadata includes dirname/revision/filename)
|
| 147 |
+
# e.g., "UD_French-GSD/r2.17/fr_gsd-ud-train.conllu" -> "fr_gsd-ud-train.conllu"
|
| 148 |
+
filename = Path(file_path).name
|
| 149 |
+
|
| 150 |
# Construct full path: UD_repos/dirname/filename
|
| 151 |
+
full_path = UD_REPOS_DIR / metadata["dirname"] / filename
|
| 152 |
|
| 153 |
if not full_path.exists():
|
| 154 |
print(f" Warning: File not found: {full_path}", file=sys.stderr)
|
|
|
|
| 186 |
"deprel": datasets.Sequence(datasets.Value("string")),
|
| 187 |
"deps": datasets.Sequence(datasets.Value("string")),
|
| 188 |
"misc": datasets.Sequence(datasets.Value("string")),
|
| 189 |
+
"mwt": [{"id": datasets.Value("string"), "form": datasets.Value("string"), "misc": datasets.Value("string")}],
|
|
|
|
|
|
|
|
|
|
|
|
|
| 190 |
})
|
| 191 |
|
| 192 |
# Create dataset from examples
|
|
|
|
| 307 |
|
| 308 |
except Exception as e:
|
| 309 |
print(f" Error: {e}", file=sys.stderr)
|
| 310 |
+
traceback.print_exc()
|
| 311 |
fail_count += 1
|
| 312 |
|
| 313 |
if verbose:
|