edumirror-repro-code / experiments /make_claim5_figure.py
ygoldi's picture
Upload folder using huggingface_hub
c5771b4 verified
Raw
History Blame Contribute Delete
2.77 kB
"""Claim 5 figure: the floor effect that makes the intervention comparison uninformative."""
import json, matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
d = json.load(open("/tmp/res/claim5/claim5.json"))
s = d["summary"]
arms = ["neglectful", "team_competition", "teacher_reminder", "pre_education"]
labels = ["Control\n(Neglectful)", "Team\nCompetition", "Teacher\nReminder", "Pre-\nEducation"]
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(11.5, 4.6))
# LEFT: the floor effect -- cooperative vastly outnumbers malicious in every arm.
coop = [s[a]["cooperative_mean"] for a in arms]
mal = [s[a]["malicious_mean"] for a in arms]
x = np.arange(len(arms)); w = 0.38
ax1.bar(x - w/2, coop, w, label="cooperative", color="#55A868")
ax1.bar(x + w/2, mal, w, label="malicious competition", color="#C44E52")
ax1.set_xticks(x); ax1.set_xticklabels(labels)
ax1.set_ylabel("Mean behaviours per episode")
ax1.set_title("The floor effect: almost no malicious competition occurred")
ax1.legend(fontsize=8); ax1.grid(axis="y", alpha=0.3)
for i, (c, m) in enumerate(zip(coop, mal)):
ax1.text(i - w/2, c + 0.3, f"{c:.1f}", ha="center", fontsize=9)
ax1.text(i + w/2, m + 0.3, f"{m:.2f}", ha="center", fontsize=9, color="#C44E52")
ax1.text(0.5, 0.62, "534 cooperative vs 11 malicious\nacross all 32 episodes\n25/32 episodes had ZERO",
transform=ax1.transAxes, fontsize=9, ha="left", va="top",
bbox=dict(boxstyle="round,pad=0.5", fc="#FFF7E0", ec="#C9A24A"))
# RIGHT: boxplot of malicious competition -- the quantity Claim 5 is about.
data = [[r["malicious_competition"] for r in d["records"] if r["arm"] == a and r["valid"]]
for a in arms]
bp = ax2.boxplot(data, tick_labels=labels, whis=(0, 100), showmeans=True, meanline=True,
meanprops={"color": "red", "linestyle": "--", "linewidth": 1.6},
medianprops={"color": "#333"}, patch_artist=True)
for patch, a in zip(bp["boxes"], arms):
patch.set_facecolor("#F2C4C4" if a == "neglectful" else "#CFD9EA")
patch.set_edgecolor("#555")
ax2.set_ylabel("Malicious competition per episode")
ax2.set_title("Paper predicts control = WIDEST spread.\nObserved: control = narrowest (range 1 vs 2)")
ax2.grid(axis="y", alpha=0.3)
ax2.set_ylim(-0.15, 2.4)
fig.suptitle("Claim 5 NOT reproduced: the control never produced the extreme competition\n"
"that the interventions are supposed to mitigate", fontsize=11)
fig.tight_layout()
fig.savefig("outputs/figures/claim5_floor_effect.png", dpi=150)
print("wrote outputs/figures/claim5_floor_effect.png")
print("coop means:", coop, "| malicious means:", mal)
print("stdevs:", [s[a]["malicious_stdev"] for a in arms])
print("ranges:", [s[a]["malicious_range"] for a in arms])