Decision score and predictions based on the nudge mechanism

RANDOM_SEED = 42
VERBOSE = False

# generate train / test data
from examples.utils.dataset import generate_train_data, generate_test_data
X_train, y_train = generate_train_data(n_normal=100, n_abnormal=20, random_seed=RANDOM_SEED)
X_test, y_test = generate_test_data(n_normal=100, n_abnormal=20, random_seed=RANDOM_SEED*2)

import matplotlib.pyplot as plt
import numpy as np
from pan import ParallelAnomalousNudge

# train PAN
model = ParallelAnomalousNudge.with_derived_estimators(X_train, y_train, nu=.1, random_seed=RANDOM_SEED, verbose=VERBOSE)
model.fit(X_train, y_train)

ParallelAnomalousNudge(estimators={np.int64(0): SomRepresentationEstimator(nu=0.1,
                                                                           random_seed=42,
                                                                           som_representation=<minisom_representation.som_representation.SomRepresentation object at 0x7fc293d660c0>),
                                   np.int64(1): SomRepresentationEstimator(nu=0.1,
                                                                           random_seed=42,
                                                                           som_representation=<minisom_representation.som_representation.SomRepresentation object at 0x7fc2a37fd820>)},
                       nu=0.1, random_seed=42, scaler=StandardScaler())

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

fig, axes = plt.subplots(2, 1, figsize=(10, 10), sharex=True, sharey=True)
axes = axes.flatten()

scores = model.score_samples(X_test)
y_values = [np.ones_like(scores[y_test == 0]), np.ones_like(scores[y_test == 1])]

ax = axes[0]
ax.set_title(rf"Nudged scores and offset calculated based on $\nu={model.nu}$", loc="left")
ax.text(0, .8, s="inlier", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="seagreen", boxstyle='round,pad=.6'), zorder=4)
ax.text(min(scores), .8, s="outlier", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="indianred", boxstyle='round,pad=.6'), zorder=4)
ax.violinplot(scores, positions=[1], vert=False)
ax.scatter(scores[y_test == 0], y=y_values[0], marker="o", edgecolor="k", color="none", s=400, alpha=.75, label="True normal", zorder=2)
ax.scatter(scores[y_test == 1], y=y_values[1], marker="x", color="crimson", s=200, alpha=.75, label="True anomaly", zorder=3)
ax.axvline(model.offset_, 0, 1, color="k", linestyle="--")
ax.text(model.offset_, 1.2, s=rf"$\text{{offset}}={model.offset_:.3}$", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="black", boxstyle='round,pad=.6'), zorder=4)
ax.legend()
ax.margins(.1)

scores = model.decision_function(X_test)
predictions = model.predict(X_test)
y_values = [np.ones_like(scores[predictions == 1]), np.ones_like(scores[predictions == -1])]

ax = axes[1]
ax.set_title(r"Decision scores (shifted nudged scores and offset)", loc="left")
ax.text(-model.offset_, .8, s="Pred 1", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="seagreen", boxstyle='round,pad=.6'), zorder=4)
ax.text(min(scores), .8, s="Pred -1", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="purple", boxstyle='round,pad=.6'), zorder=4)
ax.violinplot(scores, positions=[1], vert=False)
ax.scatter(scores[predictions == 1], y=y_values[0], marker="o", edgecolor="green", color="none", s=200, alpha=.75, label="Normal prediction", zorder=2)
ax.scatter(scores[predictions == -1], y=y_values[1], marker="s", edgecolor="purple", color="none", s=200, alpha=.75, label="Abnormal prediction", zorder=3)
ax.axvline(0, 0, 1, color="k", linestyle="--")
ax.text(0, 1.2, s=r"$\text{decision boundary}=0$", va="center", ha="center", fontsize=12, color="white", bbox=dict(color="black", boxstyle='round,pad=.6'), zorder=4)
ax.legend()
ax.margins(.1)


plt.yticks([])
plt.show()

/home/runner/work/PAN/PAN/examples/demonstrate_07_inspect_predictions.py:41: MatplotlibDeprecationWarning: vert: bool was deprecated in Matplotlib 3.11 and will be removed in 3.13. Use orientation: {'vertical', 'horizontal'} instead.
  ax.violinplot(scores, positions=[1], vert=False)
/home/runner/work/PAN/PAN/examples/demonstrate_07_inspect_predictions.py:57: MatplotlibDeprecationWarning: vert: bool was deprecated in Matplotlib 3.11 and will be removed in 3.13. Use orientation: {'vertical', 'horizontal'} instead.
  ax.violinplot(scores, positions=[1], vert=False)