scripts: add make_pie_charts.py

scripts/make_pie_charts.py

@@ -0,0 +1,129 @@
+#!/usr/bin/env python3
+"""Generate a 4-panel pie-chart summary for the dataset card.
+
+Panels:
+  (A) Image resolution — 320×240 vs 640×480
+  (B) Gel variant — markered vs markerless
+  (C) Source contribution — 8 subsets (legend instead of inline labels
+                                       to avoid overlap)
+  (D) Gel variant × source — donut: outer ring = gel variant,
+                                    inner ring = subset within variant
+"""
+import glob, os
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+import numpy as np
+import pyarrow.parquet as pq
+
+BASE = "/media/yxma/Disk1/yuxiang/mini_data_parquet"
+OUT  = f"{BASE}/assets"
+SUBSETS = ["fota_unlabeled", "real_tactile_mnist", "gelslam", "feelanyforce",
+           "threedcal", "fota_labeled", "feats", "tactile_tracking"]
+# Reordered descending by typical size so colors line up sensibly.
+
+def aggregate():
+    agg = {}
+    for sub in SUBSETS:
+        paths = sorted(glob.glob(f"{BASE}/{sub}/*.parquet"))
+        m = u = 0
+        res = {}
+        for p in paths:
+            t = pq.read_table(p, columns=["markered", "width", "height"])
+            for mki, wi, hi in zip(t.column("markered").to_pylist(),
+                                   t.column("width").to_pylist(),
+                                   t.column("height").to_pylist()):
+                if mki: m += 1
+                else: u += 1
+                res[(wi, hi)] = res.get((wi, hi), 0) + 1
+        agg[sub] = {"markered": m, "markerless": u, "res": res, "total": m + u}
+    return agg
+
+
+def main():
+    agg = aggregate()
+    fig, axes = plt.subplots(2, 2, figsize=(13, 12))
+
+    # (A) by resolution
+    tot_res = {}
+    for sub in SUBSETS:
+        for k, v in agg[sub]["res"].items():
+            tot_res[k] = tot_res.get(k, 0) + v
+    ax = axes[0, 0]
+    items = sorted(tot_res.items(), key=lambda kv: -kv[1])
+    labels = [f"{w}×{h}\n{n:,}  ({100*n/sum(tot_res.values()):.1f}%)"
+              for (w, h), n in items]
+    vals = [n for (_, _), n in items]
+    ax.pie(vals, labels=labels, colors=["#4c95d6", "#d6794c"],
+           startangle=90, wedgeprops={"edgecolor": "white", "linewidth": 2},
+           textprops={"fontsize": 12})
+    ax.set_title("Image resolution\n(GelSight Mini native modes)",
+                 fontsize=14, pad=14)
+
+    # (B) by gel variant
+    m = sum(agg[s]["markered"] for s in SUBSETS)
+    u = sum(agg[s]["markerless"] for s in SUBSETS)
+    ax = axes[0, 1]
+    ax.pie([u, m],
+           labels=[f"markerless\n{u:,}  ({100*u/(m+u):.1f}%)",
+                   f"markered\n{m:,}  ({100*m/(m+u):.1f}%)"],
+           colors=["#4c95d6", "#d6794c"], startangle=90,
+           wedgeprops={"edgecolor": "white", "linewidth": 2},
+           textprops={"fontsize": 12})
+    ax.set_title("Gel variant", fontsize=14, pad=14)
+
+    # (C) by source — legend instead of inline labels
+    ax = axes[1, 0]
+    src_vals = [agg[s]["total"] for s in SUBSETS]
+    src_total = sum(src_vals)
+    cmap = plt.cm.tab10(np.linspace(0, 1, max(len(SUBSETS), 10)))
+    wedges, _ = ax.pie(src_vals, colors=cmap, startangle=90,
+                       wedgeprops={"edgecolor": "white", "linewidth": 2})
+    legend_labels = [f"{s:<19s} {n:>7,d}  ({100*n/src_total:5.2f}%)"
+                     for s, n in zip(SUBSETS, src_vals)]
+    ax.legend(wedges, legend_labels, loc="center left",
+              bbox_to_anchor=(1.0, 0.5), fontsize=10,
+              prop={"family": "monospace", "size": 10},
+              frameon=False, title="subset (frames)",
+              title_fontsize=11)
+    ax.set_title("Source contribution", fontsize=14, pad=14)
+
+    # (D) gel variant × source donut
+    ax = axes[1, 1]
+    outer_sizes = [u, m]
+    outer_colors = ["#4c95d6", "#d6794c"]
+    ax.pie(outer_sizes,
+           labels=[f"markerless\n{u:,}", f"markered\n{m:,}"],
+           colors=outer_colors, radius=1.0,
+           wedgeprops={"edgecolor": "white", "linewidth": 2, "width": 0.35},
+           startangle=90, textprops={"fontsize": 12, "fontweight": "bold"},
+           labeldistance=1.1)
+    inner_sizes, inner_colors, inner_lbls = [], [], []
+    blue_pal = plt.cm.Blues(np.linspace(0.4, 0.9, len(SUBSETS)))
+    orng_pal = plt.cm.Oranges(np.linspace(0.4, 0.9, len(SUBSETS)))
+    for i, s in enumerate(SUBSETS):
+        if agg[s]["markerless"] > 0:
+            inner_sizes.append(agg[s]["markerless"])
+            inner_colors.append(blue_pal[i])
+            inner_lbls.append(s if agg[s]["markerless"] > u * 0.06 else "")
+    for i, s in enumerate(SUBSETS):
+        if agg[s]["markered"] > 0:
+            inner_sizes.append(agg[s]["markered"])
+            inner_colors.append(orng_pal[i])
+            inner_lbls.append(s if agg[s]["markered"] > m * 0.06 else "")
+    ax.pie(inner_sizes, radius=0.62, colors=inner_colors,
+           wedgeprops={"edgecolor": "white", "linewidth": 1, "width": 0.30},
+           startangle=90, labels=inner_lbls, labeldistance=0.78,
+           textprops={"fontsize": 8})
+    ax.set_title("Gel variant × source (donut)", fontsize=14, pad=14)
+
+    fig.suptitle(f"gelsight-mini-pretrain · summary pie charts  "
+                 f"(total {m+u:,} frames)", fontsize=16, y=0.995)
+    plt.tight_layout(rect=[0, 0, 1, 0.97])
+    out = f"{OUT}/summary_pies.png"
+    plt.savefig(out, dpi=140, bbox_inches="tight")
+    print(f"saved {out}  total={m+u:,}  markered={m:,}  markerless={u:,}")
+
+
+if __name__ == "__main__":
+    main()