plot_utils.py

import matplotlib.pyplot as plt
import numpy as np


def plot_histories(
        history,
        name: str,
        n_epochs=10,
):
    """
    Plot history of the passed models.
    :param history: History of the first model.
    :param name: Name of the first model.
    :param n_epochs: Number of epochs to plot.
    """
    # Get the n first values of the metrics
    acc_key, loss_key, val_acc_key, val_loss_key = 'accuracy', 'loss', 'val_accuracy', 'val_loss'
    if not all(k in history.history for k in [acc_key, loss_key, val_acc_key, val_loss_key]):
        print(f'Fehlende Metriken in History von {name}')
        return
    tr_acc: np.ndarray = _first_n(history, acc_key, n_epochs)
    tr_loss: np.ndarray = _first_n(history, loss_key, n_epochs)
    va_acc: np.ndarray = _first_n(history, val_acc_key, n_epochs)
    va_loss: np.ndarray = _first_n(history, val_loss_key, n_epochs)

    # Create figures
    fig, axes = plt.subplots(1, 2, figsize=(12, 5))
    epochs = np.arange(1, len(tr_loss) + 1)

    # Plot loss
    axes[0].plot(epochs, tr_loss, label=f"{name} train loss")
    axes[0].plot(epochs, va_loss, label=f"{name} val loss")
    axes[0].set_xlabel("Epoche")
    axes[0].set_ylabel("Loss")
    axes[0].set_title(f"Loss (erste {len(epochs)} Epochen)")
    axes[0].legend()
    axes[0].grid(True)

    # Plot accuracy
    axes[1].plot(epochs, tr_acc, label=f"{name} train acc")
    axes[1].plot(epochs, va_acc, label=f"{name} val acc")
    axes[1].set_xlabel("Epoche")
    axes[1].set_ylabel("Accuracy")
    axes[1].set_title(f"Accuracy (erste {len(epochs)} Epochen)")
    axes[1].legend()
    axes[1].grid(True)

    plt.tight_layout()
    plt.show()

def _first_n(history, key: str, n: int = 10) -> np.ndarray:
    """
    Get the n first values of a metric.
    :param history: History of the model.
    :key: Metric to get.
    :n: Number of values to get.
    """
    return np.array(history.history[key][:n], dtype=float)