IKMLab/MPTR_AutoT

1from pathlib import Path2import torch3import os4import random5import argparse6import json7import pandas as pd8import numpy as np9from sklearn.metrics import precision_recall_fscore_support10from ast import literal_eval11 12 13def pred_by_threshold(14    t: float,15    y_true: np.array,16    similarities: np.array,17    classes: dict,18):19    preds = (similarities >= t) * 120    sk_results = precision_recall_fscore_support(21        y_true,22        preds,23        # average="samples",  # For calculating sample-wise P and R scores.24    )25    outputs = {26        "f1": np.average(sk_results[2]),27        "P": np.average(sk_results[0]),28        "R": np.average(sk_results[1]),29    }30    for label_name, idx in classes.items():31        outputs[f"{label_name}_f1"] = sk_results[2][idx]32    return outputs33 34 35def get_avg_length(dataset: torch.utils.data.Dataset):36    all_lengths = 037    data_size = len(dataset)38    for i in range(data_size):39        all_lengths += len(dataset[i]["input_ids"])40    return all_lengths / data_size41 42 43def load_csv_multi_label(filename: str, col_name: str = "labels") -> pd.DataFrame:44    """Prevent Pandas from converting lists of int into lists of strings.45 46    Args:47        filename (str): path of a csv file48        col_name (str, optional): column name of lists of int. Defaults to 'labels'.49 50    Returns:51        pd.DataFrame: a Pandas dataframe52    """53    return pd.read_csv(filename, converters={col_name: literal_eval})54 55 56def save_logged_results(filename: str, results: dict):57    try:58        old_df = pd.read_csv(filename)59        df = pd.concat([old_df, pd.DataFrame(results)], ignore_index=True)60    except FileNotFoundError:61        df = pd.DataFrame(results)62 63    df.to_csv(filename, index=None)64 65 66def set_seed(seed):67    """68    Args:69        seed: an integer number to initialize a pseudorandom number generator70    """71    os.environ["PYTHONHASHSEED"] = str(seed)72    random.seed(seed)73    np.random.seed(seed)74    torch.manual_seed(seed)75 76    if torch.cuda.is_available():77        torch.cuda.manual_seed(seed)78        # torch.cuda.manual_seed_all(seed)  # if using more than one GPUs79        torch.backends.cudnn.deterministic = True80        torch.backends.cudnn.benchmark = False81 82 83def save_baseline_table(84    y_preds: list,85    baseline_name: str,86    baseline_result_file: str = "results/baselines.pkl",87    all_doc_idx: list = None,88) -> None:89    if Path(baseline_result_file).exists():90        df = pd.read_pickle(baseline_result_file)91    else:92        assert all_doc_idx is not None93        df = pd.DataFrame({"doc_idx": all_doc_idx})94 95    df[baseline_name] = y_preds96    df.to_pickle(baseline_result_file)97 98 99def load_params(path_of_params):100    with open(path_of_params, "r") as f:101        params = json.load(f)102    return argparse.Namespace(**params)103 104 105def get_label_words(classes: list, use_multi_label_words=False) -> list:106    mapping = {107        "cyst": "cyst",108        "HCC": "hcc",  # hepatoma109        "cirrhosis": "cirrhosis",110        "post-treatment": "posttreatment",111        "steatosis": "steatosis",112        "metastasis": "metastasis",113        "hemangioma": "hemangioma",114    }115    if use_multi_label_words:116        mapping = {117            "cyst": ["cyst"],118            "HCC": ["hcc", "hepatoma"],  # hepatoma119            "cirrhosis": ["cirrhosis"],120            "post-treatment": ["posttreatment"],121            "steatosis": ["steatosis", "steatohepatitis"],122            "metastasis": ["metastasis"],123            "hemangioma": ["hemangioma"],124        }125 126    label_words = [mapping[c] for c in classes]127    return label_words128 129 130def seed_mapper(data_type: str) -> list:131    mapping = {132        "train_8": [2, 4, 7, 11, 21, 23, 24, 36, 44, 128],133        "train_32": [0, 1, 3, 7, 10],134    }135    if data_type in mapping:136        return mapping[data_type]137    else:138        raise NotImplementedError139