Upload preprocessing/filter_ids.py with huggingface_hub

preprocessing/filter_ids.py

@@ -0,0 +1,34 @@
+import argparse
+import os
+
+import numpy as np
+from datasets import load_from_disk
+
+from const import used_indices
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("input_path")
+    parser.add_argument("--output_path")
+    return parser.parse_args()
+
+
+def filter_ids(rows):
+    return {
+        "sentence1_ling": np.array(rows["sentence1_lftk+"])[:, used_indices],
+        "sentence2_ling": np.array(rows["sentence2_lftk+"])[:, used_indices],
+    }
+
+
+def main():
+    args = parse_args()
+    output_path = args.output_path or args.input_path.rstrip("/") + "_filtered"
+
+    data = load_from_disk(args.input_path)
+    data = data.map(filter_ids, batched=True, batch_size=1000, num_proc=min(8, os.cpu_count() or 1))
+    data.save_to_disk(output_path)
+
+
+if __name__ == "__main__":
+    main()