linear -> poly

model/model.pickle

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:73bf71607b6b99d8576a79ec96cdf97e008134e7d348477f93b8cdcf057db19e
-size 3411728
+oid sha256:b2dc963399b56d847a6b5e93ab807d18d1dbc8b89285b5b151dc17e6608ef72d
+size 3182918

train.py

@@ -54,7 +54,7 @@ def save_confusion_matrix(y: NDArray, pred: NDArray):
 
 def main() -> None:
     train_df = pl.read_parquet(DATA / "train.parquet")
-    clf = SVC(kernel="linear", probability=True)
+    clf = SVC(kernel="poly", probability=True)
     clf.fit(
         train_df.get_column("embeds").to_numpy(),
         train_df.get_column("is_news").to_numpy(),

usage.py

@@ -18,11 +18,12 @@ def get_model():
 
 
 def get_record():
-    df = pl.read_parquet(DATA / "eval.parquet")
+    df = pl.read_csv(DATA / "eval.csv")
+    raw = {col: val for col, val in zip(df.columns, df.sample().row(0))}
+    text_fields = ["meta_title", "meta_description", "content"]
     return {
-        col: val
-        for col, val in zip(df.columns, df.sample().row(0))
-        if col in ["text", "is_news"]
+        "text": "\n\n".join(raw[k] for k in text_fields),
+        "is_news": raw["is_news_article"],
     }