Log predictions to patonlab/analytics dataset

requirements.txt

@@ -21,4 +21,7 @@ scipy==1.3.3
 scikit-learn==0.21.3
 tqdm==4.64.1
 
+# HF dataset logging
+huggingface_hub==0.16.4
+
 # nfp is vendored locally — no PyPI entry needed

worker.py

@@ -12,6 +12,7 @@ warnings.filterwarnings("ignore", category=FutureWarning)
 import json
 import math
 import pickle
+import datetime
 from io import StringIO
 
 import redis
@@ -68,6 +69,46 @@ redis_client = redis.StrictRedis(
     host="localhost", port=6379, db=0, decode_responses=True
 )
 
+# ── Analytics logging to HF Dataset ──────────────────────────────────────────
+_HF_TOKEN       = os.environ.get("HF_TOKEN", "")
+_ANALYTICS_REPO = "patonlab/analytics"
+_ANALYTICS_FILE = "data.csv"
+
+def _log_prediction():
+    """Append one row to the existing patonlab/analytics data.csv.
+    Format matches the alfabet log: space,timestamp
+    """
+    if not _HF_TOKEN:
+        return
+    try:
+        from huggingface_hub import HfApi
+        import tempfile
+
+        api = HfApi(token=_HF_TOKEN)
+        timestamp = datetime.datetime.utcnow().isoformat()
+
+        # Download the current CSV, append a row, re-upload
+        with tempfile.TemporaryDirectory() as tmpdir:
+            local_path = os.path.join(tmpdir, "data.csv")
+            api.hf_hub_download(
+                repo_id=_ANALYTICS_REPO,
+                filename=_ANALYTICS_FILE,
+                repo_type="dataset",
+                local_dir=tmpdir,
+            )
+            with open(local_path, "a") as f:
+                f.write(f"patonlab/cascade,{timestamp}\n")
+
+            api.upload_file(
+                path_or_fileobj=local_path,
+                path_in_repo=_ANALYTICS_FILE,
+                repo_id=_ANALYTICS_REPO,
+                repo_type="dataset",
+                commit_message=f"log: cascade prediction {timestamp[:10]}",
+            )
+    except Exception as e:
+        print(f"Analytics logging failed (non-fatal): {e}", flush=True)
+
 
 def _mol_to_sdf(mol, conf_id=0):
     sio = StringIO()
@@ -140,7 +181,6 @@ def _fmt_relative_E(spread_df, energy_order):
 def run_job(task_id, smiles, type_):
     result_key = f"task_result_{task_id}"
     try:
-        # Prepare mol — preprocess_C/H expects a mol without Hs (it adds them internally)
         mol = Chem.MolFromSmiles(smiles)
         AllChem.EmbedMolecule(mol, useRandomCoords=True)
         mol_with_h = Chem.AddHs(mol, addCoords=True)
@@ -149,12 +189,8 @@ def run_job(task_id, smiles, type_):
         mol_with_confs, ids, nr = genConf(mol_with_h, rms=-1, nc=200, efilter=10.0, rmspost=0.5)
         print(f"genConf: {len(ids)} conformers", flush=True)
 
-        # Build SDFs directly from genConf mol (energy order, lowest first)
         conf_sdfs, energy_order = _build_sdfs_from_genconf(mol_with_confs, ids)
 
-        # Run GNN preprocessing and inference using the same mol_with_confs
-        # Pass mol_with_confs as a list — preprocess will use its existing conformers
-        # by calling genConf internally, but we suppress that output
         mols = [Chem.MolFromSmiles(smiles)]
         for m in mols:
             AllChem.EmbedMolecule(m, useRandomCoords=True)
@@ -215,6 +251,9 @@ def run_job(task_id, smiles, type_):
         redis_client.set(result_key, json.dumps(result), ex=3600)
         print(f"Task {task_id} complete — {len(conf_sdfs)} conformers", flush=True)
 
+        # Log to analytics dataset (non-blocking)
+        _log_prediction()
+
     except Exception as e:
         import traceback; traceback.print_exc()
         redis_client.set(result_key, json.dumps({"errMessage": str(e)}), ex=3600)