src/autotrain/trainers/image_classification/utils.py [10:122]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - BINARY_CLASSIFICATION_EVAL_METRICS = ( "eval_loss", "eval_accuracy", "eval_f1", "eval_auc", "eval_precision", "eval_recall", ) MULTI_CLASS_CLASSIFICATION_EVAL_METRICS = ( "eval_loss", "eval_accuracy", "eval_f1_macro", "eval_f1_micro", "eval_f1_weighted", "eval_precision_macro", "eval_precision_micro", "eval_precision_weighted", "eval_recall_macro", "eval_recall_micro", "eval_recall_weighted", ) MODEL_CARD = """ --- tags: - autotrain - transformers - image-classification{base_model} widget: - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/tiger.jpg example_title: Tiger - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/teapot.jpg example_title: Teapot - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/palace.jpg example_title: Palace{dataset_tag} --- # Model Trained Using AutoTrain - Problem type: Image Classification ## Validation Metrics {validation_metrics} """ def _binary_classification_metrics(pred): """ Computes various binary classification metrics given the predictions and labels. Args: pred (tuple): A tuple containing raw predictions and true labels. raw_predictions (numpy.ndarray): The raw prediction scores from the model. labels (numpy.ndarray): The true labels. Returns: dict: A dictionary containing the following metrics: - f1 (float): The F1 score. - precision (float): The precision score. - recall (float): The recall score. - auc (float): The Area Under the ROC Curve (AUC) score. - accuracy (float): The accuracy score. """ raw_predictions, labels = pred predictions = np.argmax(raw_predictions, axis=1) result = { "f1": metrics.f1_score(labels, predictions), "precision": metrics.precision_score(labels, predictions), "recall": metrics.recall_score(labels, predictions), "auc": metrics.roc_auc_score(labels, raw_predictions[:, 1]), "accuracy": metrics.accuracy_score(labels, predictions), } return result def _multi_class_classification_metrics(pred): """ Compute various classification metrics for multi-class classification. Args: pred (tuple): A tuple containing raw predictions and true labels. - raw_predictions (numpy.ndarray): The raw prediction scores for each class. - labels (numpy.ndarray): The true labels. Returns: dict: A dictionary containing the following metrics: - "f1_macro": F1 score with macro averaging. - "f1_micro": F1 score with micro averaging. - "f1_weighted": F1 score with weighted averaging. - "precision_macro": Precision score with macro averaging. - "precision_micro": Precision score with micro averaging. - "precision_weighted": Precision score with weighted averaging. - "recall_macro": Recall score with macro averaging. - "recall_micro": Recall score with micro averaging. - "recall_weighted": Recall score with weighted averaging. - "accuracy": Accuracy score. """ raw_predictions, labels = pred predictions = np.argmax(raw_predictions, axis=1) results = { "f1_macro": metrics.f1_score(labels, predictions, average="macro"), "f1_micro": metrics.f1_score(labels, predictions, average="micro"), "f1_weighted": metrics.f1_score(labels, predictions, average="weighted"), "precision_macro": metrics.precision_score(labels, predictions, average="macro"), "precision_micro": metrics.precision_score(labels, predictions, average="micro"), "precision_weighted": metrics.precision_score(labels, predictions, average="weighted"), "recall_macro": metrics.recall_score(labels, predictions, average="macro"), "recall_micro": metrics.recall_score(labels, predictions, average="micro"), "recall_weighted": metrics.recall_score(labels, predictions, average="weighted"), "accuracy": metrics.accuracy_score(labels, predictions), } return results - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - src/autotrain/trainers/text_classification/utils.py [8:115]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - BINARY_CLASSIFICATION_EVAL_METRICS = ( "eval_loss", "eval_accuracy", "eval_f1", "eval_auc", "eval_precision", "eval_recall", ) MULTI_CLASS_CLASSIFICATION_EVAL_METRICS = ( "eval_loss", "eval_accuracy", "eval_f1_macro", "eval_f1_micro", "eval_f1_weighted", "eval_precision_macro", "eval_precision_micro", "eval_precision_weighted", "eval_recall_macro", "eval_recall_micro", "eval_recall_weighted", ) MODEL_CARD = """ --- library_name: transformers tags: - autotrain - text-classification{base_model} widget: - text: "I love AutoTrain"{dataset_tag} --- # Model Trained Using AutoTrain - Problem type: Text Classification ## Validation Metrics {validation_metrics} """ def _binary_classification_metrics(pred): """ Calculate various binary classification metrics. Args: pred (tuple): A tuple containing raw predictions and true labels. - raw_predictions (numpy.ndarray): The raw prediction scores from the model. - labels (numpy.ndarray): The true labels. Returns: dict: A dictionary containing the following metrics: - "f1" (float): The F1 score. - "precision" (float): The precision score. - "recall" (float): The recall score. - "auc" (float): The Area Under the ROC Curve (AUC) score. - "accuracy" (float): The accuracy score. """ raw_predictions, labels = pred predictions = np.argmax(raw_predictions, axis=1) result = { "f1": metrics.f1_score(labels, predictions), "precision": metrics.precision_score(labels, predictions), "recall": metrics.recall_score(labels, predictions), "auc": metrics.roc_auc_score(labels, raw_predictions[:, 1]), "accuracy": metrics.accuracy_score(labels, predictions), } return result def _multi_class_classification_metrics(pred): """ Compute various classification metrics for multi-class classification. Args: pred (tuple): A tuple containing raw predictions and true labels. - raw_predictions (numpy.ndarray): The raw prediction scores for each class. - labels (numpy.ndarray): The true labels. Returns: dict: A dictionary containing the following metrics: - "f1_macro": F1 score with macro averaging. - "f1_micro": F1 score with micro averaging. - "f1_weighted": F1 score with weighted averaging. - "precision_macro": Precision score with macro averaging. - "precision_micro": Precision score with micro averaging. - "precision_weighted": Precision score with weighted averaging. - "recall_macro": Recall score with macro averaging. - "recall_micro": Recall score with micro averaging. - "recall_weighted": Recall score with weighted averaging. - "accuracy": Accuracy score. """ raw_predictions, labels = pred predictions = np.argmax(raw_predictions, axis=1) results = { "f1_macro": metrics.f1_score(labels, predictions, average="macro"), "f1_micro": metrics.f1_score(labels, predictions, average="micro"), "f1_weighted": metrics.f1_score(labels, predictions, average="weighted"), "precision_macro": metrics.precision_score(labels, predictions, average="macro"), "precision_micro": metrics.precision_score(labels, predictions, average="micro"), "precision_weighted": metrics.precision_score(labels, predictions, average="weighted"), "recall_macro": metrics.recall_score(labels, predictions, average="macro"), "recall_micro": metrics.recall_score(labels, predictions, average="micro"), "recall_weighted": metrics.recall_score(labels, predictions, average="weighted"), "accuracy": metrics.accuracy_score(labels, predictions), } return results - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -