Utils

---

Helper functions

predict_and_report(network, gen, save_path='./', csv_save_path=None, report=True, save_df=True, batch_size=32, input_shape=None)

Inference the network on given dataset and save predictions in csv file. If needed, a separate report with f1-score, precision and recall for each class will be saved.

Parameters:

Name	Type	Description	Default
network	tf.keras.Model	network to predict with	required
gen	src.datasets.datagen.ImageDatagen	dataset generator. Will be used to infer dataset, features and filenames	required
save_path	str	root directory to save the prediction file in	'./'
report	bool	whether to save separate classification report. For more info, check sklearn.metrics.classification_report	True
save_df	bool	whether to save a dataframe containing predictions. It will be filtered by the number of mispredictions (from best to worst)	True
input_shape	Optional[List[int]]	input shape of the network	None
batch_size	int	batch size to use during inference	32
csv_save_path	str = None	save path for csv. If not provided, will be inferred from save_path directory	None

Source code in conftrainer/utils.py

def predict_and_report(network: Any, gen: Any, save_path: str = "./",
                       csv_save_path: Optional[str] = None, report: bool = True,
                       save_df: bool = True, batch_size: int = 32,
                       input_shape: Optional[List[int]] = None) -> None:
    """
    Inference the network on given dataset and save predictions in csv file. If needed,
    a separate report with f1-score, precision and recall for each class will be saved.

    Parameters
    ----------
    network : tf.keras.Model
        network to predict with
    gen : src.datasets.datagen.ImageDatagen
        dataset generator. Will be used to infer dataset, features and filenames
    save_path : str
        root directory to save the prediction file in
    report : bool, default: True
        whether to save separate classification report. For more info,
        check sklearn.metrics.classification_report
    save_df : bool, default: True
        whether to save a dataframe containing predictions. It will be filtered by the number of
        mispredictions (from best to worst)
    input_shape : Optional[List[int]]
        input shape of the network
    batch_size : int, default: 32
        batch size to use during inference
    csv_save_path : str = None
        save path for csv. If not provided, will be inferred from save_path directory
    """
    dataset = gen.create_ds(shape=input_shape, batch_size=batch_size, training=False)
    dest_path = os.path.join(save_path, gen.name + "_predictions")
    y_prob = network.predict(dataset, verbose=1)
    is_multibranch = isinstance(y_prob, list)  # Regular network outputs a np.ndarray
    y_true = gen.labels
    y_true_processed = gen.probs_to_labels(y_true)
    y_pred = gen.probs_to_labels(y_prob)
    if report and (y_true is not None):
        if is_multibranch:
            report = {}
            for labels, preds, classes, name in zip(y_true_processed, y_pred, gen.classes, gen.task_names):
                report[name] = classification_report(labels,
                                                     preds,
                                                     target_names=classes,
                                                     output_dict=True
                                                     )

        else:
            report = classification_report(y_true_processed,
                                           y_pred,
                                           target_names=gen.classes,
                                           output_dict=True
                                           )
        report_path = dest_path + "_report.json"
        with open(report_path, "w+", encoding="utf-8") as file:
            json.dump(report, file)

    if save_df:
        filenames = np.array(gen.filepaths)
        col_names = gen.classes
        if is_multibranch:
            col_names = [name for task_classes in col_names for name in task_classes] if col_names else None
            y_prob = np.concatenate(y_prob, axis=1)
        df_pred = pd.DataFrame(y_prob, columns=col_names)
        df_pred["Path"] = filenames
        if not csv_save_path:
            csv_save_path = dest_path + ".csv"
        df_pred.to_csv(csv_save_path, index=False)

save_yaml(obj, filepath, default_flow_style=False)

Save the dict-like object into YAML file

Parameters:

Name	Type	Description	Default
obj	object	object to save	required
filepath	str	path to the file to create	required
default_flow_style	bool	corresponding parameter for yaml.dump. If True, nested dicts will be written in JSON style (A: {key: value, ...}) , otherwise block style will be used:	False

Source code in conftrainer/utils.py

def save_yaml(obj: object, filepath: str, default_flow_style: bool = False) -> None:
    """
    Save the dict-like object into YAML file

    Parameters
    ----------
    obj : object
        object to save
    filepath : str
        path to the file to create
    default_flow_style : bool
        corresponding parameter for yaml.dump. If True, nested dicts will be written in JSON
        style (A: {key: value, ...}) , otherwise block style will be used:
    """
    with open(filepath, "w+", encoding="utf-8") as file:
        yaml.dump(obj, file, default_flow_style=default_flow_style, sort_keys=False)

json_load(path, encoding='utf-8')

Load json file

Parameters:

Name	Type	Description	Default
path	str	path of json file	required
encoding	str	encoding of json file	'utf-8'

Returns:

Name	Type	Description
obj	Dict | List	loaded json file

Source code in conftrainer/utils.py

def json_load(path: str, encoding: str = 'utf-8') -> Union[Dict, List]:
    """
    Load json file

    Parameters
    ----------
    path : str
        path of json file
    encoding : str
        encoding of json file

    Returns
    -------
    obj : Dict | List
        loaded json file
    """
    with open(path, 'r', encoding=encoding) as file:
        obj = json.load(file)
        return obj

json_save(path, obj)

Save dict to json

Parameters:

Name	Type	Description	Default
path	str	path of json file	required
obj	Dict	input	required

Source code in conftrainer/utils.py

def json_save(path: str, obj: Union[Dict, List[Dict]]) -> None:
    """
    Save dict to json

    Parameters
    ----------
    path : str
        path of json file
    obj : Dict
        input
    Returns
    -------
    """
    with open(path, 'w', encoding='utf-8') as file:
        json.dump(obj, file)

plot_class_metrics(y_true, y_pred, class_names)

Plot a heatmap showing f1 score, precision and recall for each class

Parameters:

Name	Type	Description	Default
y_true	np.ndarray	true and predicted labels	required
y_pred	np.ndarray	true and predicted labels	required
class_names	List[str]	names of classes to display	required

Returns:

Name	Type	Description
out	Figure	heatmap with classwise metrix

Source code in conftrainer/utils.py

def plot_class_metrics(y_true: np.ndarray, y_pred: np.ndarray, class_names: List[str]) -> Figure:
    """
    Plot a heatmap showing f1 score, precision and recall for each class

    Parameters
    ----------
    y_true, y_pred : np.ndarray
        true and predicted labels
    class_names : List[str]
        names of classes to display

    Returns
    -------
    out : Figure
        heatmap with classwise metrix
    """
    precision = precision_score(y_true=y_true, y_pred=y_pred, average=None)
    recall = recall_score(y_true=y_true, y_pred=y_pred, average=None)
    f1 = f1_score(y_true=y_true, y_pred=y_pred, average=None)
    inds = np.argsort(f1)
    metrics = np.stack([f1[inds], precision[inds], recall[inds]], axis=0).round(3)
    metric_names = ["f1_score", "precision", "recall"]
    fig, ax = plt.subplots(figsize=(12, len(class_names)))

    ax.imshow(metrics.T, cmap="Wistia")

    # Show all ticks and label them with the respective list entries
    ax.set_xticks(np.arange(len(metric_names)), labels=metric_names)
    ax.set_yticks(np.arange(len(class_names)), labels=np.array(class_names)[inds])

    # Rotate the tick labels and set their alignment.
    plt.setp(ax.get_xticklabels(), rotation=45, ha="right", rotation_mode="anchor")

    # Loop over data dimensions and create text annotations.
    for i in range(len(class_names)):
        for j in range(len(metrics)):
            ax.text(j, i, metrics[j, i],
                    ha="center", va="center", color="b")
    return fig

create_object(config, modules)

Create an object based on given config. First the function import the class from given modules, then initializes an object with given params. Note that if several modules have the same class, first one will be imported and initialized.

Parameters:

Name	Type	Description	Default
config	ObjInitConfig	configuration to create an object. Contains name and arguments	required
modules	List	modules to attempt an import from	required

Returns:

Name	Type	Description
out		initialized object

Source code in conftrainer/utils.py

def create_object(config: ObjInitConfig, modules: List):
    """
    Create an object based on given config. First the function import the class from given
    modules, then initializes an object with given params. Note that if several modules have the
    same class, first one will be imported and initialized.

    Parameters
    ----------
    config : ObjInitConfig
        configuration to create an object. Contains name and arguments
    modules : List
        modules to attempt an import from

    Returns
    -------
    out:
        initialized object
    """
    if not modules:
        raise ValueError("Please provide a non-empty list of modules to attempt import from")

    for module in modules:
        obj_class = getattr(module, config.name, None)
        if obj_class:
            return obj_class(**config.args)
    module_names = [module.__name__ for module in modules]
    raise ValueError(f"Please provide valid class from {module_names}. Current config: {config}")

plot_confusion_matrix(y_true, y_pred, class_names=None, **kwargs)

Calculate and plot confusion matrix

Parameters:

Name	Type	Description	Default
y_true	np.ndarray	true labels	required
y_pred	np.ndarray	predictions of the network	required
class_names	List[str]	names of classes to display on the plot	None

Returns:

Name	Type	Description
out	matplotlib.figure.Figure	a plot containing confusion matrix

Source code in conftrainer/utils.py

def plot_confusion_matrix(y_true: np.ndarray, y_pred: np.ndarray, class_names: List[str] = None, **kwargs) -> Figure:
    """
    Calculate and plot confusion matrix

    Parameters
    ----------
    y_true : np.ndarray
        true labels
    y_pred : np.ndarray
        predictions of the network
    class_names : List[str]
        names of classes to display on the plot

    Returns
    -------
    out : matplotlib.figure.Figure
        a plot containing confusion matrix
    """
    matrix = confusion_matrix(y_true=y_true,
                              y_pred=y_pred,
                              **kwargs)
    matrix = np.round(matrix, 2)
    fig, ax = plt.subplots(1, 1, figsize=(16, 12))
    display = ConfusionMatrixDisplay(matrix, display_labels=class_names)
    display.plot(ax=ax, xticks_rotation="vertical")
    return fig