Visualisation¶

torchtsetlin.viz needs matplotlib (pip install "torchtsetlin[viz]"). Every function returns the Axes it drew on.

Function	What it shows
`plot_history(history)`	metric curves per epoch
`plot_confusion_matrix(cm, class_names)`	confusion matrix heat-map
`plot_clause_memory(model, clause)`	book-style memory of one clause: bar per literal, dashed include boundary
`plot_ta_states(model)`	heat-map of all automata (clauses × literals)
`plot_literal_frequency(model)`	how many clauses include each literal
`plot_feature_importance(importance, names)`	horizontal bars
`plot_conv_clause(model, clause)` / `plot_conv_clauses(model, clauses)`	convolutional clause patches (black = on, white = off, gray = free) with the allowed region; 3D clauses are drawn slice by slice (`depth=` picks one)
`plot_votes(votes, y)`	distribution of winning vote sums, correct vs incorrect
`plot_trustworthiness(curve)`	accuracy and coverage vs confidence
`plot_clause_weights(model)`	weights of weighted / coalesced models

import matplotlib.pyplot as plt
fig, axes = plt.subplots(1, 2, figsize=(12, 4))
tt.viz.plot_clause_memory(model, clause=0, ax=axes[0])
tt.viz.plot_votes(model(x_test), y_test, ax=axes[1])
plt.show()