Upload convert_rlcd.py with huggingface_hub

convert_rlcd.py

@@ -0,0 +1,63 @@
+import hashlib, json, random
+from huggingface_hub import hf_hub_download
+
+random.seed(42)
+
+
+def conv(path, cap):
+    rows = []
+    for line in open(path):
+        r = json.loads(line)
+        qs, tgts = [], []
+        for qid, q in r["questions"].items():
+            t = q["type"]
+            if t == "noul":
+                p = float(q["reference"])
+                if not 0.0 <= p <= 1.0:
+                    continue
+                qs.append({"kind": "noul", "options": ["false", "true"],
+                           "text": q["instructions"], "unknown_allowed": False})
+                tgts.append([1.0 - p, p])
+            elif t == "choice":
+                crit = q["criteria"]
+                keys = list(crit.keys())
+                ref = q["reference"]
+                if any(k not in ref for k in keys) or len(keys) < 2:
+                    continue
+                qs.append({"kind": "choice", "options": [f"{k}: {crit[k]}" for k in keys],
+                           "text": q["instructions"], "unknown_allowed": False})
+                tgts.append([float(ref[k]) for k in keys])
+            elif t == "score":
+                crit, ref = q["criteria"], q["reference"]
+                if len(crit) != len(ref):
+                    continue
+                qs.append({"kind": "score", "options": [f"{i} - {c}" for i, c in enumerate(crit)],
+                           "text": q["instructions"], "unknown_allowed": False})
+                tgts.append([float(x) for x in ref])
+        if not qs:
+            continue
+        st = r["state"] if isinstance(r["state"], str) else json.dumps(r["state"], sort_keys=True)
+        gid = hashlib.sha256(st.encode()).hexdigest()[:16]
+        rows.append({"state": st, "id": f"rlcd:{r.get('domain','?')}:{gid}:{len(rows)}",
+                     "questions": qs, "targets": tgts,
+                     "source": f"rlcd:{r.get('domain','?')}", "group_id": f"rlcd:{gid}"})
+        if len(rows) >= cap:
+            break
+    return rows
+
+
+tr = conv(hf_hub_download("soyrsoyr/openjev-rlcd-v0", "train.jsonl", repo_type="dataset"), 26000)
+dv = conv(hf_hub_download("soyrsoyr/openjev-rlcd-v0", "val.jsonl", repo_type="dataset"), 4000)
+trg = {r["group_id"] for r in tr}
+dv = [r for r in dv if r["group_id"] not in trg]
+random.shuffle(tr)
+with open("/root/alea/data/rlcd_train.jsonl", "w") as f:
+    for r in tr:
+        f.write(json.dumps(r) + "\n")
+with open("/root/alea/data/rlcd_dev.jsonl", "w") as f:
+    for r in dv:
+        f.write(json.dumps(r) + "\n")
+print({"train": len(tr), "dev": len(dv),
+       "train_qs": sum(len(r["questions"]) for r in tr),
+       "kinds": {k: sum(1 for r in tr for q in r["questions"] if q["kind"] == k)
+                 for k in ("choice", "noul", "score")}})