| import pandas as pd | |
| from sklearn.model_selection import StratifiedKFold | |
| from data_loader.load_data_ import load_data | |
| def stratified_kfold(cfg): | |
| whole_df = load_data('/opt/ml/input/data/train/train.tsv') | |
| skf = StratifiedKFold(n_splits=cfg.values.val_args.num_k, | |
| shuffle=True, | |
| random_state=cfg.values.seed) | |
| for k, (_, val_idx) in enumerate(skf.split(X=whole_df, y=whole_df['label'].values)): | |
| whole_df.loc[val_idx, 'kfold'] = int(k) | |
| whole_df.to_csv('/opt/ml/input/data/train/train_folds.tsv', index=False) | |
| print('Divde K folds Done!!!') | |
| # def hp_search(trial: optuna.Trial, | |
| # model_name: str, | |
| # dataset, | |
| # label_nbr, | |
| # metric_name, | |
| # device): | |
| # """ | |
| # objective function for optuna.study optimizes for epoch number, lr and batch_size | |
| # :param trial: optuna.Trial, trial of optuna, which will optimize for hyperparameters | |
| # :param model_name: name of the pretrained model | |
| # :param dataset: huggingface/nlp dataset object | |
| # :param label_nbr: number of label for the model to output | |
| # :param metric_name: name of the metric to maximize | |
| # :param reference_class: reference class to calculate metrics for | |
| # :param device: device where the training will occur, cuda recommended | |
| # :return: metric after training | |
| # """ | |
| # lr = trial.suggest_float("lr", 1e-7, 1e-4, log=True) | |
| # batch_size = trial.suggest_categorical("batch_size", [2, 4, 6]) | |
| # epochs = trial.suggest_int("epochs", 3, 6) | |
| # model = MultilabeledSequenceModel(pretrained_model_name=model_name, | |
| # label_nbr=label_nbr).to(device) | |
| # optimizer = AdamW(params=model.parameters(), lr=lr) | |
| # for epoch in range(epochs): | |
| # train_epoch(model, | |
| # optimizer, | |
| # dataset, | |
| # batch_size, | |
| # device) | |
| # labels, preds = evaluate(model, | |
| # dataset, | |
| # batch_size, | |
| # device) | |
| # metric = accuracy_score(labels, preds) | |
| # trial.report(metric, epoch) | |
| # if trial.should_prune(): | |
| # raise optuna.TrialPruned() | |
| # return metric | |
| # def train_epoch(model, | |
| # optimizer, | |
| # dataset, | |
| # batch_size, | |
| # device): | |
| # # trains a model for an epoch, creating a dataloader from a huggingface/nlp dataset | |
| # # the parameters are auto-explanatory | |
| # # assumes model already on device | |
| # dataloader_train = DataLoader(dataset['train'], | |
| # shuffle=True, | |
| # batch_size=batch_size) | |
| # for batch in tqdm(dataloader_train, total=len(dataloader_train)): | |
| # optimizer.zero_grad() | |
| # preds = model(batch['input_ids'].long().to(device)) | |
| # loss = F.cross_entropy(preds, batch['labels'].to(device)) | |
| # loss.backward() | |
| # optimizer.step() | |
| # def evaluate(model, | |
| # dataset, | |
| # batch_size, | |
| # device): | |
| # # evaluates a model by getting the predictions, aside with labels, of a dataset | |
| # # creates the dataloader from a huggingface/nlp dataset | |
| # # assumes model already in device | |
| # dataloader_test = DataLoader(dataset['valid'], | |
| # shuffle=True, | |
| # batch_size=batch_size) | |
| # with torch.no_grad(): | |
| # eval_preds = [] | |
| # eval_labels = [] | |
| # for batch in tqdm(dataloader_test, total=len(dataloader_test)): | |
| # preds = model(batch['input_ids'].long().to(device)) | |
| # preds = preds.argmax(dim=-1) | |
| # eval_preds.append(preds.cpu().numpy()) | |
| # eval_labels.append(batch['labels'].cpu().numpy()) | |
| # return np.concatenate(eval_labels), np.concatenate(eval_preds) | |