Coverage for source/plotting/classification_testing_plot_responsibility_chain.py: 97%
106 statements
« prev ^ index » next coverage.py v7.8.0, created at 2025-08-23 15:31 +0000
1# plotting/classification_testing_plot_responsibility_chain.py
3# global imports
4import logging
5import matplotlib.pyplot as plt
6import numpy as np
7from matplotlib.gridspec import GridSpec
8from sklearn.metrics import RocCurveDisplay
10# local imports
11from source.agent import ClassificationTestingStrategyHandler
12from source.plotting import PlotResponsibilityChainBase
14class ClassificationTestingPlotResponsibilityChain(PlotResponsibilityChainBase):
15 """
16 Implements a plotting responsibility chain for classification testing results.
17 It implements the _can_plot and _plot methods to visualize confusion matrices,
18 classification reports, and ROC curves.
19 """
21 # local constants
22 __ADDITIONAL_REPORT_LABELS = ["accuracy", "macro avg", "weighted avg"]
24 def _can_plot(self, key: str) -> bool:
25 """
26 Checks if the plot can be generated for the given key.
28 Parameters:
29 key (str): The key to check.
31 Returns:
32 (bool): True if the plot can be generated, False otherwise.
33 """
35 return key == ClassificationTestingStrategyHandler.PLOTTING_KEY
37 def _plot(self, plot_data: dict) -> plt.Axes:
38 """
39 Generates the classification testing plot based on the provided data.
41 Parameters:
42 plot_data (dict): The data to be plotted.
44 Returns:
45 (plt.Axes): The axes object containing the plot.
46 """
48 conf_matrix = plot_data.get("confusion_matrix", None)
49 class_report = plot_data.get("classification_report", None)
50 prediction_probabilities = plot_data.get("prediction_probabilities", None)
51 true_labels = plot_data.get("true_labels", None)
53 if conf_matrix is None or class_report is None or prediction_probabilities is None or true_labels is None:
54 logging.warning(f"Insufficient data for plotting results under key: {ClassificationTestingStrategyHandler.PLOTTING_KEY}.")
55 plt.text(0.5, 0.5, "Insufficient data for plotting",
56 ha = 'center', va = 'center', fontsize = 12)
57 return plt.gca()
59 additional_report = {}
60 for additional_label in self.__ADDITIONAL_REPORT_LABELS:
61 if additional_label in class_report:
62 additional_report[additional_label] = class_report.pop(additional_label)
64 fig = plt.figure(figsize = self._EXPECTED_FIGURE_SIZE)
65 gs = GridSpec(2, 2, figure = fig)
66 classes = list(class_report.keys())
67 shortened_classes_names = [class_name[:3] for class_name in classes]
69 # Plot 1: Confusion Matrix as a heatmap
70 ax1 = plt.subplot(gs[0, 0])
71 ax1.set_title(f"Confusion Matrix (Accuracy: {additional_report['accuracy']:.2%})")
73 normalized_conf_matrix = conf_matrix.astype('float') / conf_matrix.sum(axis = 1, keepdims = True)
74 normalized_conf_matrix = np.round(np.nan_to_num(normalized_conf_matrix, nan = 0.0), 2)
75 ax1.imshow(normalized_conf_matrix, interpolation = 'nearest', cmap = plt.cm.GnBu)
77 tick_marks = np.arange(len(classes))
78 ax1.set_xticks(tick_marks)
79 ax1.set_yticks(tick_marks)
80 ax1.set_xticklabels(shortened_classes_names)
81 ax1.set_yticklabels(shortened_classes_names)
82 ax1.set_xlabel('Predicted label')
83 ax1.set_ylabel('True label')
85 for i in range(conf_matrix.shape[0]):
86 for j in range(conf_matrix.shape[1]):
87 color = "white" if normalized_conf_matrix[i, j] > 0.5 else "black"
88 ax1.text(j, i - 0.1, format(conf_matrix[i, j], 'd'),
89 ha = "center", va = "center", fontsize = 10, weight = 'bold', color = color)
90 ax1.text(j, i + 0.15, f'{normalized_conf_matrix[i, j]:.2f}',
91 ha = "center", va = "center", fontsize = 8, color = color)
93 # Plot 2: Precision, Recall, F1 Score Bar Chart
94 ax2 = plt.subplot(gs[1, 0])
95 precision_scores = []
96 recall_scores = []
97 f1_scores = []
99 for metrics_dict in class_report.values():
100 precision_scores.append(metrics_dict["precision"])
101 recall_scores.append(metrics_dict["recall"])
102 f1_scores.append(metrics_dict["f1-score"])
104 shift = 0.2
105 precision_bars = ax2.bar(tick_marks - shift, precision_scores, shift, label = 'Precision')
106 recall_bars = ax2.bar(tick_marks, recall_scores, shift, label = 'Recall')
107 f1_bars = ax2.bar(tick_marks + shift, f1_scores, shift, label = 'F1-score')
109 for i, (precision_bar, recall_bar, f1_bar) in enumerate(zip(precision_bars, recall_bars, f1_bars)):
110 ax2.text(precision_bar.get_x() + (precision_bar.get_width() / 2),
111 precision_bar.get_height() + 0.01 if precision_bar.get_height() < 0.9 else \
112 precision_bar.get_height() - 0.01, f'{precision_scores[i]:.3f}',
113 ha = 'center', va = 'bottom' if precision_bar.get_height() < 0.9 else 'top', rotation = 90,
114 fontsize = 8, weight = 'bold')
115 ax2.text(recall_bar.get_x() + (recall_bar.get_width() / 2),
116 recall_bar.get_height() + 0.01 if recall_bar.get_height() < 0.9 else \
117 recall_bar.get_height() - 0.01, f'{recall_scores[i]:.3f}',
118 ha = 'center', va = 'bottom' if recall_bar.get_height() < 0.9 else 'top', rotation = 90,
119 fontsize = 8, weight = 'bold')
120 ax2.text(f1_bar.get_x() + (f1_bar.get_width() / 2),
121 f1_bar.get_height() + 0.01 if f1_bar.get_height() < 0.9 else \
122 f1_bar.get_height() - 0.01, f'{f1_scores[i]:.3f}',
123 ha = 'center', va = 'bottom' if f1_bar.get_height() < 0.9 else 'top', rotation = 90,
124 fontsize = 8, weight = 'bold')
126 ax2.set_title('Classification metrics by class')
127 ax2.set_xticks(tick_marks)
128 ax2.set_xticklabels(shortened_classes_names)
129 ax2.set_xlabel('Classes')
130 ax2.set_ylabel('Score')
131 ax2.set_ylim([0, 1])
132 ax2.legend(fontsize = 'x-small')
134 # Plot 3: OvR-ROC curves
135 ax3 = plt.subplot(gs[0, 1])
137 for i, class_name in enumerate(classes):
138 y_true_class_binary = (true_labels == i).astype(int)
139 y_score = prediction_probabilities[:, i]
140 RocCurveDisplay.from_predictions(y_true_class_binary, y_score, name = f"{class_name}",
141 ax = ax3, plot_chance_level = (i == len(classes) - 1))
143 ax3.set_title('One-vs-Rest ROC curves')
144 ax3.set_xlabel('False positive rate')
145 ax3.set_ylabel('True positive rate')
146 ax3.grid(alpha = 0.3)
147 ax3.legend(loc = "lower right", fontsize = 'x-small')
148 plt.tight_layout()
150 # Plot 4: Macro avg and weighted avg
151 ax4 = plt.subplot(gs[1, 1])
152 additional_labels = list(additional_report.keys())[1:]
153 precision_scores = []
154 recall_scores = []
155 f1_scores = []
157 for metrics in additional_report.values():
158 if isinstance(metrics, dict):
159 precision_scores.append(metrics['precision'])
160 recall_scores.append(metrics['recall'])
161 f1_scores.append(metrics['f1-score'])
163 x = np.arange(len(additional_labels))
164 precision_bars = ax4.bar(x - shift, precision_scores, shift, label = 'Precision')
165 recall_bars = ax4.bar(x, recall_scores, shift, label = 'Recall')
166 f1_bars = ax4.bar(x + shift, f1_scores, shift, label = 'F1-score')
168 for i, (precision_bar, recall_bar, f1_bar) in enumerate(zip(precision_bars, recall_bars, f1_bars)):
169 ax4.text(precision_bar.get_x() + (precision_bar.get_width() / 2),
170 precision_bar.get_height() + 0.01 if precision_bar.get_height() < 0.9 else \
171 precision_bar.get_height() - 0.01, f'{precision_scores[i]:.3f}',
172 ha = 'center', va = 'bottom' if precision_bar.get_height() < 0.9 else 'top', rotation = 90,
173 fontsize = 8, weight = 'bold')
174 ax4.text(recall_bar.get_x() + (recall_bar.get_width() / 2),
175 recall_bar.get_height() + 0.01 if recall_bar.get_height() < 0.9 else \
176 recall_bar.get_height() - 0.01, f'{recall_scores[i]:.3f}',
177 ha = 'center', va = 'bottom' if recall_bar.get_height() < 0.9 else 'top', rotation = 90,
178 fontsize = 8, weight = 'bold')
179 ax4.text(f1_bar.get_x() + (f1_bar.get_width() / 2),
180 f1_bar.get_height() + 0.01 if f1_bar.get_height() < 0.9 else \
181 f1_bar.get_height() - 0.01, f'{f1_scores[i]:.3f}',
182 ha = 'center', va = 'bottom' if f1_bar.get_height() < 0.9 else 'top', rotation = 90,
183 fontsize = 8, weight = 'bold')
185 ax4.set_title('Macro avg and weighted avg')
186 ax4.set_xticks(x)
187 ax4.set_xticklabels(additional_labels)
188 ax4.set_xlabel('Metrics')
189 ax4.set_ylabel('Score')
190 ax4.set_ylim([0, 1])
191 ax4.legend(fontsize = 'x-small')
192 plt.tight_layout()
194 return plt.gca()