src/asr/build_dataframe.py

import json, glob, os
import pandas as pd
from typing import Dict, List

def _read_rows(results_dir: str) -> pd.DataFrame:
    rows = []
    for path in glob.glob(os.path.join(results_dir, "*.json")):
        with open(path, "r") as f:
            data = json.load(f)
        for _, rec in data.items():
            rows.append({
                "Model": rec.get("model"),
                "Corpus": rec.get("corpus"),
                "Split": rec.get("split"),
                "WER": rec.get("wer"),
                "CER": rec.get("cer"),
                "Domain": rec.get("domain", ""),  # 'in' or 'out' or ''
                "Evaluated": rec.get("evaluated_at"),
            })
    return pd.DataFrame(rows)

def per_dataset_tables(results_dir: str) -> Dict[str, pd.DataFrame]:
    """Return a dict of {corpus_name: DataFrame(Model, WER, CER)}"""
    df = _read_rows(results_dir)
    if df.empty:
        return {}
    tables = {}
    for corpus, g in df.groupby("Corpus"):
        t = (
            g.sort_values("WER", ascending=True)
             .loc[:, ["Model", "WER", "CER"]]
             .reset_index(drop=True)
        )
        tables[corpus] = t
    return tables

def overall_table(results_dir: str) -> pd.DataFrame:
    """
    Return a single table with per-model averages:
      - Avg WER (all corpora)
      - Avg CER (all corpora)
      - Avg WER (excl. in-domain)
      - Avg CER (excl. in-domain)
    """
    df = _read_rows(results_dir)
    if df.empty:
        return pd.DataFrame()

    # compute overall averages
    agg_all = (
        df.groupby("Model")[["WER", "CER"]]
          .mean()
          .rename(columns={"WER": "Avg WER (all)", "CER": "Avg CER (all)"})
    )

    # compute averages excluding in-domain rows
    df_out = df[df["Domain"] != "in"]
    if df_out.empty:
        agg_out = pd.DataFrame(index=agg_all.index, columns=["Avg WER (excl in)", "Avg CER (excl in)"])
    else:
        agg_out = (
            df_out.groupby("Model")[["WER", "CER"]]
                  .mean()
                  .rename(columns={"WER": "Avg WER (excl in)", "CER": "Avg CER (excl in)"})
        )

    # merge and sort by best average WER (all)
    merged = pd.concat([agg_all, agg_out], axis=1)
    merged = merged.reset_index()
    if "Avg WER (all)" in merged.columns:
        merged = merged.sort_values("Avg WER (all)", ascending=True)
    return merged

def list_corpora(results_dir: str) -> List[str]:
    df = _read_rows(results_dir)
    if df.empty:
        return []
    return sorted(df["Corpus"].unique().tolist())