kopia lustrzana https://github.com/thinkst/zippy
91 wiersze
2.9 KiB
Python
91 wiersze
2.9 KiB
Python
#!/usr/bin/env python3
|
|
|
|
import numpy as np
|
|
import matplotlib.pyplot as plt
|
|
from sklearn.metrics import roc_curve, auc
|
|
import re
|
|
from junitparser import JUnitXml
|
|
|
|
MODELS = ['zippy-lzma', 'contentatscale', 'roberta', 'gptzero', 'crossplag']
|
|
SKIPCASES = ['gpt2', 'gpt3']
|
|
|
|
MAX_PER_CASE = 500
|
|
|
|
plt.figure()
|
|
|
|
for model in MODELS:
|
|
xml = JUnitXml.fromfile(f'test_results/{model}-report.xml')
|
|
cases = []
|
|
for suite in xml:
|
|
for case in suite:
|
|
cases.append(case)
|
|
|
|
truths = []
|
|
scores = []
|
|
per_case = {}
|
|
fails_per_case = {}
|
|
for c in cases:
|
|
if c.name is None:
|
|
print("ERROR")
|
|
continue
|
|
cname = re.sub('\[.*$', '', c.name)
|
|
if any(sub in cname for sub in SKIPCASES):
|
|
continue
|
|
if cname in per_case.keys():
|
|
per_case[cname] += 1
|
|
else:
|
|
per_case[cname] = 1
|
|
fails_per_case[cname] = 0
|
|
if per_case[cname] > MAX_PER_CASE:
|
|
continue
|
|
try:
|
|
score = float(c._elem.getchildren()[0].getchildren()[0].values()[1])
|
|
except:
|
|
continue
|
|
if 'human' in c.name:
|
|
truths.append(1)
|
|
if c.is_passed:
|
|
scores.append(score)
|
|
else:
|
|
fails_per_case[cname] += 1
|
|
scores.append(score * -1.0)
|
|
else:
|
|
truths.append(-1)
|
|
if c.is_passed:
|
|
scores.append(score * -1.0)
|
|
else:
|
|
fails_per_case[cname] += 1
|
|
scores.append(score)
|
|
|
|
y_true = np.array(truths)
|
|
y_scores = np.array(scores)
|
|
print("Failures per case for " + model)
|
|
print(fails_per_case)
|
|
tf = 0
|
|
for k in fails_per_case.keys():
|
|
tf += fails_per_case[k]
|
|
print('Total fails: ' + str(tf))
|
|
tp = len(cases) - tf
|
|
# Compute the false positive rate (FPR), true positive rate (TPR), and threshold values
|
|
fpr, tpr, thresholds = roc_curve(y_true, y_scores)
|
|
gmeans = np.sqrt(tpr * (1-fpr))
|
|
ix = np.argmax(gmeans)
|
|
print('Best Threshold=%f, G-Mean=%.3f' % (thresholds[ix], gmeans[ix]))
|
|
#print(thresholds)
|
|
# calculate the g-mean for each threshold
|
|
# locate the index of the largest g-mean
|
|
# Calculate the area under the ROC curve (AUC)
|
|
roc_auc = auc(fpr, tpr)
|
|
|
|
# Plot the ROC curve
|
|
plt.plot(fpr, tpr, lw=2, label=f'{model.capitalize()}: ROC curve (%Acc = {tp/len(cases):0.2f}; AUC = {roc_auc:0.2f})')
|
|
plt.scatter(fpr[ix], tpr[ix], marker='o', color='black')#, label=model.capitalize() + ': Best @ threshold = %0.2f' % thresholds[ix])
|
|
|
|
plt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--', label="Random classifier")
|
|
plt.xlim([0.0, 1.0])
|
|
plt.ylim([0.0, 1.05])
|
|
plt.xlabel('False Positive Rate')
|
|
plt.ylabel('True Positive Rate')
|
|
plt.title('Receiver Operating Characteristic for LLM detection')
|
|
plt.legend(loc="lower right")
|
|
plt.savefig('ai_detect_roc.png') |