src/fuse_esdac/x_01_fuse.py

import argparse
import json
from pathlib import Path

import pandas as pd

from utils import fuse_csv, fuse_tif

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--checkpoint", type=str, required=True,
                        help="Checkpoint config")
    parser.add_argument("--last-only", action="store_true",
                        help="Only do the last dataset (for debugging)")
    parser.add_argument("--recover_unfinished", action="store_true",
                        help="Recover from samples_unfinished.json and skip finished datasets")
    args = parser.parse_args()

    # -------------------------
    # Load checkpoint config
    # -------------------------
    with open(f"./src/fuse_esdac/configs/{args.checkpoint}.json", "r") as f:
        cfg = json.load(f)

    input_json = Path(f"./src/fuse_esdac/outputs/{cfg.get('input')}/samples.json")
    datasets = cfg.get("datasets", [])
    if args.last_only and datasets:
        datasets = datasets[-1:]

    # -------------------------
    # Prepare output dir (needed for unfinished save/recovery)
    # -------------------------
    out_dir = Path(f"./src/fuse_esdac/outputs/{args.checkpoint}")
    out_dir.mkdir(parents=True, exist_ok=True)

    out_json = out_dir / f"samples.json"
    out_var_csv = out_dir / f"stat_variable.csv"
    unfinished_json = out_dir / "samples_unfinished.json"
    finished_datasets_json = out_dir / "datasets_finished.json"

    # -------------------------
    # Load previous fused dict (normal) OR recover unfinished
    # -------------------------
    if args.recover_unfinished:
        if not unfinished_json.exists():
            raise FileNotFoundError(f"--recover_unfinished requires {unfinished_json} to exist")
        with open(unfinished_json, "r") as f:
            big_output_dict = json.load(f)

        if finished_datasets_json.exists():
            with open(finished_datasets_json, "r") as f:
                finished_datasets = set(json.load(f))
        else:
            finished_datasets = set()

        # Skip finished ones (preserve order)
        datasets = [d for d in datasets if d not in finished_datasets]
    else:
        finished_datasets = set()
        if input_json and Path(input_json).exists():
            with open(input_json, "r") as f:
                big_output_dict = json.load(f)
        else:
            big_output_dict = {}

    # -------------------------
    # Fuse datasets
    # -------------------------
    for dataset in datasets:
        try:
            data_dir = Path(f"./datasets/esdac/{dataset}/processed")
            schema_path = Path(f"./src/esdac/{dataset}/fuse_schema.json")
            with open(schema_path, "r") as f:
                schema = json.load(f)

            for filename in schema.keys():
                if filename.endswith(".csv"):
                    csv_path = data_dir / filename
                    big_output_dict = fuse_csv(
                        source_file=csv_path,
                        schema=schema[filename],
                        source_name=dataset,
                        current_dict=big_output_dict,
                    )
                elif filename.endswith("@tif"):
                    tif_path = data_dir / filename[:-4]
                    big_output_dict = fuse_tif(
                        source_file=tif_path,
                        schema=schema[filename],
                        source_name=dataset,
                        current_dict=big_output_dict,
                    )

            # Mark dataset finished (for recovery skipping)
            finished_datasets.add(dataset)
            with open(finished_datasets_json, "w") as f:
                json.dump(sorted(finished_datasets), f, indent=2, ensure_ascii=False)

        except Exception:
            # Save partial progress and re-raise
            with open(unfinished_json, "w") as f:
                json.dump(big_output_dict, f, indent=2, ensure_ascii=False)
            raise

    # -------------------------
    # Save outputs
    # -------------------------
    with open(out_json, "w") as f:
        json.dump(big_output_dict, f, indent=2, ensure_ascii=False)

    # -------------------------
    # Variable-level table
    # -------------------------
    table = {}
    for sample_key, sample in big_output_dict.items():
        prefix = sample_key.split("_")[0]
        if prefix not in table:
            table[prefix] = {}
        for domain, section in sample.items():
            if isinstance(section, dict):
                for var, leaf in section.items():
                    if isinstance(leaf, dict):
                        if leaf.get("value") not in (None, "", "null"):
                            key = f"{domain}:{var}"
                            table[prefix][key] = table[prefix].get(key, 0) + 1

    # Count raw samples per prefix (independent of filled variables)
    sample_count = {}
    for sample_key in big_output_dict.keys():
        prefix = sample_key.split("_")[0]
        sample_count[prefix] = sample_count.get(prefix, 0) + 1

    df_var = (
        pd.DataFrame.from_dict(table, orient="index")
        .fillna(0)
        .infer_objects()
        .astype(int)
    )

    # Ensure all prefixes are present even if they had no filled variables
    all_prefixes = sorted(sample_count.keys())
    if not df_var.empty:
        df_var = df_var.reindex(all_prefixes, fill_value=0)
    else:
        df_var = pd.DataFrame(index=all_prefixes)

    # Add true sample counts per row
    df_var["TOTAL_SAMPLES"] = df_var.index.map(lambda p: sample_count.get(p, 0)).astype(int)

    # Add TOTAL row with correct sums and total sample count
    totals_no_samples_col = df_var.drop(columns=["TOTAL_SAMPLES"]).sum(axis=0).astype(int)
    df_var.loc["SUM"] = 0  # initialize the row
    df_var.loc["SUM", totals_no_samples_col.index] = totals_no_samples_col
    df_var.loc["SUM", "TOTAL_SAMPLES"] = int(sum(sample_count.values()))
    df_var = df_var.astype(int)

    df_var.to_csv(out_var_csv)