metric="minmax",
feature_type="words",
sample_authors=False,
sample_features=False,
n_features=n_features,
n_test_pairs=n_pairs,
n_dev_pairs=n_pairs,
em_iterations=100,
vector_space_model="std",
weight=0.2,
eps=0.01,
norm="l2",
balanced_pairs=False)
logging.info("Starting verification [dev / test]")
verifier.vectorize(X_dev, X_test)
dev_results, test_results = verifier.predict()
logging.info("Computing results")
# first prec rec curve of test results:
test_Fs, test_Ps, test_Rs, test_Ts = evaluate(test_results)
fig = sb.plt.figure()
sb.plt.xlabel("recall", fontsize=10)
sb.plt.ylabel("precision", fontsize=10)
sb.plt.xlim(0.4, 1)
sb.plt.ylim(0.4, 1.05)
sb.plt.plot(test_Rs, test_Ps, label="baseline")
# get max for dev:
dev_Fs, dev_Ps, dev_Rs, dev_Ts = evaluate(test_results)
best_t = dev_Ts[np.nanargmax(dev_Fs)]
baseline_test_f, test_p, test_r = evaluate_with_threshold(test_results, t=best_t)
print dm
print vsm
verifier = Verification(random_state=1000,
metric=dm,
n_features=10000,
n_dev_pairs=0,
n_test_pairs=99999999,
vector_space_model=vsm,
balanced_pairs=False,
control_pairs=False)
logging.info("Starting verification [train / test]")
verifier.vectorize(X_dev, X_test)
train_results, test_results = verifier.predict(filter_imposters=False)
logging.info("Computing results")
test_df = verifier.get_distance_table(verifier.test_dists, verifier.test_pairs,
"test")
test_df.to_csv("../outputs/caesar_test.csv")
test_df = pd.read_csv("../outputs/caesar_test.csv")
test_df = test_df.set_index("id")
test_df = test_df.applymap(lambda x: int(x * 1000)).corr()
# heatmap plotting:
sb.heatmap(test_df)
ax = sb.plt.gca()
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
label.set_fontname('Arial')
metric="minmax",
feature_type="words",
sample_authors=False,
sample_features=False,
n_features=n_features,
n_test_pairs=n_pairs,
n_dev_pairs=n_pairs,
em_iterations=100,
vector_space_model="std",
weight=0.2,
eps=0.01,
norm="l2",
balanced_pairs=False)
logging.info("Starting verification [dev / test]")
verifier.vectorize(X_dev, X_test)
dev_results, test_results = verifier.predict()
logging.info("Computing results")
# first prec rec curve of test results:
test_Fs, test_Ps, test_Rs, test_Ts = evaluate(test_results)
fig = sb.plt.figure()
sb.plt.xlabel("recall", fontsize=10)
sb.plt.ylabel("precision", fontsize=10)
sb.plt.xlim(0.4, 1)
sb.plt.ylim(0.4, 1.05)
sb.plt.plot(test_Rs, test_Ps, label="baseline")
# get max for dev:
dev_Fs, dev_Ps, dev_Rs, dev_Ts = evaluate(test_results)
best_t = dev_Ts[np.nanargmax(dev_Fs)]
baseline_test_f, test_p, test_r = evaluate_with_threshold(test_results,
print dm
print vsm
verifier = Verification(random_state=1000,
metric=dm,
n_features=10000,
n_dev_pairs=0,
n_test_pairs=99999999,
vector_space_model=vsm,
balanced_pairs=False,
control_pairs=False)
logging.info("Starting verification [train / test]")
verifier.vectorize(X_dev, X_test)
train_results, test_results = verifier.predict(filter_imposters=False)
logging.info("Computing results")
test_df = verifier.get_distance_table(verifier.test_dists, verifier.test_pairs, "test")
test_df.to_csv("../outputs/caesar_test.csv")
test_df = pd.read_csv("../outputs/caesar_test.csv")
test_df = test_df.set_index("id")
test_df = test_df.applymap(lambda x:int(x*1000)).corr()
# heatmap plotting:
sb.heatmap(test_df)
ax = sb.plt.gca()
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
label.set_fontname('Arial')
label.set_fontsize(3)