Add some plots

.gitignore

@@ -4,4 +4,3 @@ original
 # Generated data
 resources/gen
 *.pyc
-scripts

scripts/dsbuild.py

@@ -0,0 +1,56 @@
+import gzip
+import json
+import os
+
+from matplotlib import pyplot as plt
+
+
+def load_stats(path: str) -> dict:
+    stats: dict = {}
+    for filename in os.listdir(path):
+        file_path: str = os.path.join(path, filename)
+        if os.path.isdir(filename):
+            continue
+        data: dict
+        if filename.endswith('.gz'):
+            with gzip.open(file_path, mode='rt') as file:
+                data = json.loads(file.read())
+        else:
+            with open(file_path, mode='rt') as file:
+                data = json.loads(file.read())
+        print(f'Loaded stats from {file_path}')
+        stats.update(**data)
+    return stats
+
+
+def stat_filter(stats: dict, deviation_cutoff=(1.0, 0.0), clamp=(200.0, 2048.0), min_messages=4) -> list[dict]:
+    cutoff_threshold: (float, float) = (
+        stats['wordsStdDev'] * deviation_cutoff[0], stats['wordsStdDev'] * deviation_cutoff[1])
+    if cutoff_threshold[1] <= 0:
+        cutoff_threshold = (cutoff_threshold[0], stats['wordsMax'])
+    cutoff_min: float = max(max(clamp[0], cutoff_threshold[0]), stats['wordsMean'] - cutoff_threshold[0])
+    cutoff_max: float = stats['wordsMean'] + cutoff_threshold[1]
+    if clamp[1] > 0:
+        cutoff_max = min(clamp[1], cutoff_max)
+
+    conversations: list[dict] = [v for k, v in stats['conversations'].items() if
+                                 v['wordsMax'] <= cutoff_max and v['wordsMin'] >= cutoff_min and v[
+                                     'messagesCount'] >= min_messages]
+    print(
+        f'Min: {cutoff_min:0.0f}\tMax: {cutoff_max:0.0f}\n'
+        f'Clamped from {cutoff_threshold[0]:0.0f}, {cutoff_threshold[1]:0.0f}')
+    print(f'{len(conversations)} conversations')
+
+    return conversations
+
+
+def build_mean_word_plot(conv_stats: list[float], title: str = 'Conversation Message Mean Words', xlabel: str = 'Mean',
+                         ylabel: str = 'Conversations', text: str = '',
+                         **kwargs):
+    fig, ax = plt.subplots(figsize=(5, 2.7), layout='constrained')
+    n, bins, patches = ax.hist(conv_stats, density=True,
+                               facecolor='C0', alpha=0.75, **kwargs)
+    ax.set_title(title)
+    ax.set_xlabel(xlabel)
+    ax.set_ylabel(ylabel)
+    plt.figtext(0, 0.95, text)

scripts/generate_plot.py

@@ -0,0 +1,31 @@
+import os
+
+import matplotlib.pyplot as plt
+
+import dsbuild
+
+stats_path: list[str] = ['../stats/full/test/', '../stats/full/train']
+
+
+def __plot_mean_word_counts(path: list[str], show: bool = False, save: bool = False):
+    for p in path:
+        p: str = os.path.normpath(p)
+        split_name: str = p.split(os.sep)[-1]
+        config_name: str = p.split(os.sep)[-2]
+
+        stats: dict = dsbuild.load_stats(p)
+        conversations: list[float] = [conversation['wordsMean'] for _, conversation in stats['conversations'].items()]
+        dsbuild.build_mean_word_plot(conversations, f'Mean Words ({config_name}-{split_name})',
+                                     text=f'{len(conversations)} total')
+        if save:
+            plt.savefig(f'../assets/{config_name}-{split_name}.png')
+        if show:
+            plt.show()
+
+
+def main():
+    __plot_mean_word_counts(stats_path, show=True, save=True)
+
+
+if __name__ == '__main__':
+    main()