| """¿Ayuda agregar poco texto sintético a datos reales? Proporciones 0, 10, 25, 50, 100% de los positivos reales. 5 semillas.""" |
| import glob, json |
| import numpy as np, pandas as pd |
| from sklearn.feature_extraction.text import TfidfVectorizer |
| from sklearn.linear_model import LogisticRegression |
| from sklearn.metrics import f1_score |
|
|
| tr = pd.read_parquet("data/piuba_real__train.parquet"); te_all = pd.read_parquet("data/piuba_real__test.parquet") |
| syn = {"racismo": pd.read_parquet("racismo/v1.1/completo.parquet"), "clasismo": pd.read_parquet("clasismo/v1.1/completo.parquet")} |
| m = pd.concat([pd.read_parquet(f) for f in glob.glob("data/machismodetect__*.parquet")]) |
| syn["machismo"] = pd.DataFrame({"texto": m.texto_machista, "etiqueta": 1}).drop_duplicates("texto") |
| G = {"racismo": "RACISM", "clasismo": "CLASS", "machismo": "WOMEN"} |
| out = {} |
| for name, g in G.items(): |
| pos = tr[tr[g] == 1]; neg = tr[tr.HATEFUL == 0].sample(8000, random_state=0) |
| te = pd.concat([te_all[te_all[g] == 1], te_all[te_all.HATEFUL == 0].sample(3000, random_state=0)]) |
| y = (te[g] == 1).astype(int) |
| sp = syn[name][syn[name].etiqueta == 1].texto |
| res = {} |
| for ratio in [0, 0.1, 0.25, 0.5, 1.0]: |
| f1s = [] |
| for seed in range(5): |
| add = sp.sample(int(ratio * len(pos)), random_state=seed) if ratio else pd.Series([], dtype=str) |
| X = pd.concat([pos.text, add, neg.text]); Y = np.r_[np.ones(len(pos) + len(add)), np.zeros(len(neg))] |
| v = TfidfVectorizer(ngram_range=(1, 2), min_df=2, sublinear_tf=True) |
| clf = LogisticRegression(max_iter=3000, class_weight="balanced").fit(v.fit_transform(X), Y) |
| f1s.append(f1_score(y, clf.predict(v.transform(te.text)))) |
| res[f"{int(ratio*100)}%"] = {"f1_media": round(float(np.mean(f1s)), 3), "f1_de": round(float(np.std(f1s)), 3)} |
| out[name] = res |
| print(name, res, flush=True) |
| json.dump(out, open("resultados_mezcla.json", "w"), ensure_ascii=False, indent=2) |
|
|
