y_pred_proba=y_pred_proba,
labels=labels,
title="Test set (Deep prediction)",
path=os.path.join(EXP_DIR, 'test_deep.pdf'))
# ====== make a streamline classifier ====== #
# training PLDA
Z3_train, y_train = make_dnn_prediction(f_z3, X=train, title="TRAIN")
print("Z3_train:", Z3_train.shape, y_train.shape)
Z3_valid, y_valid = make_dnn_prediction(f_z3, X=valid, title="VALID")
print("Z3_valid:", Z3_valid.shape, y_valid.shape)
plda = PLDA(n_phi=200,
random_state=K.get_rng().randint(10e8),
n_iter=12,
labels=labels,
verbose=0)
plda.fit(np.concatenate([Z3_train, Z3_valid], axis=0),
np.concatenate([y_train, y_valid], axis=0))
y_pred_log_proba = plda.predict_log_proba(Z3_test)
evaluate(y_true=X_test_true,
y_pred_log_proba=y_pred_log_proba,
labels=labels,
title="Test set (PLDA - Latent prediction)",
path=os.path.join(EXP_DIR, 'test_latent.pdf'))
# ====== visualize ====== #
visualize_latent_space(X_org=X_test_data,
X_latent=Z1_test,
name=X_test_name,
labels=X_test_true,
title="latent1")
visualize_latent_space(X_org=X_test_data,
X_latent=Z2_test,
name=X_test_name,
# ====== tsne ====== #
tsne = TSNE(n_components=NUM_DIM,
perplexity=30.0,
learning_rate=200.0,
n_iter=1000,
random_state=SEED)
X_train_tsne = tsne.fit_transform(X_train)
X_score_tsne = tsne.fit_transform(X_score)
# ====== lda ====== #
lda = LinearDiscriminantAnalysis(n_components=NUM_DIM)
lda.fit(X_train, y_train)
X_train_lda = lda.transform(X_train)
X_score_lda = lda.transform(X_score)
# ====== plda ====== #
plda = PLDA(n_phi=NUM_DIM, random_state=SEED)
plda.fit(X_train, y_train)
X_train_plda = plda.predict_log_proba(X_train)
X_score_plda = plda.predict_log_proba(X_score)
# ====== gmm ====== #
gmm = GaussianMixture(n_components=NUM_DIM,
max_iter=100,
covariance_type='full',
random_state=SEED)
gmm.fit(X_train)
X_train_gmm = gmm._estimate_weighted_log_prob(X_train)
X_score_gmm = gmm._estimate_weighted_log_prob(X_score)
# ====== rbm ====== #
rbm = BernoulliRBM(n_components=NUM_DIM,
batch_size=8,
learning_rate=0.0008,
n_iter=8,
# ===========================================================================
y_pred_proba, Z1_test, Z2_test, Z3_test = make_dnn_prediction(
functions=[f_pred_proba, f_z1, f_z2, f_z3], X=X_test_data, title='TEST')
print("Test Latent:", Z1_test.shape, Z2_test.shape, Z3_test.shape)
y_pred = np.argmax(y_pred_proba, axis=-1)
evaluate(y_true=X_test_true, y_pred_proba=y_pred_proba, labels=labels,
title="Test set (Deep prediction)",
path=os.path.join(EXP_DIR, 'test_deep.pdf'))
# ====== make a streamline classifier ====== #
# training PLDA
Z3_train, y_train = make_dnn_prediction(f_z3, X=train, title="TRAIN")
print("Z3_train:", Z3_train.shape, y_train.shape)
Z3_valid, y_valid = make_dnn_prediction(f_z3, X=valid, title="VALID")
print("Z3_valid:", Z3_valid.shape, y_valid.shape)
plda = PLDA(n_phi=200, random_state=K.get_rng().randint(10e8),
n_iter=12, labels=labels, verbose=0)
plda.fit(np.concatenate([Z3_train, Z3_valid], axis=0),
np.concatenate([y_train, y_valid], axis=0))
y_pred_log_proba = plda.predict_log_proba(Z3_test)
evaluate(y_true=X_test_true, y_pred_log_proba=y_pred_log_proba, labels=labels,
title="Test set (PLDA - Latent prediction)",
path=os.path.join(EXP_DIR, 'test_latent.pdf'))
# ====== visualize ====== #
visualize_latent_space(X_org=X_test_data, X_latent=Z1_test,
name=X_test_name, labels=X_test_true,
title="latent1")
visualize_latent_space(X_org=X_test_data, X_latent=Z2_test,
name=X_test_name, labels=X_test_true,
title="latent2")
V.plot_save(os.path.join(EXP_DIR, 'latent.pdf'))
random_state=SEED)
X_train_tsne_pca = tsne_pca.fit_transform(X_train_pca)
X_score_tsne_pca = tsne_pca.fit_transform(X_score_pca)
# ====== tsne ====== #
tsne = TSNE(n_components=NUM_DIM, perplexity=30.0, learning_rate=200.0, n_iter=1000,
random_state=SEED)
X_train_tsne = tsne.fit_transform(X_train)
X_score_tsne = tsne.fit_transform(X_score)
# ====== lda ====== #
lda = LinearDiscriminantAnalysis(n_components=NUM_DIM)
lda.fit(X_train, y_train)
X_train_lda = lda.transform(X_train)
X_score_lda = lda.transform(X_score)
# ====== plda ====== #
plda = PLDA(n_phi=NUM_DIM, random_state=SEED)
plda.fit(X_train, y_train)
X_train_plda = plda.predict_log_proba(X_train)
X_score_plda = plda.predict_log_proba(X_score)
# ====== gmm ====== #
gmm = GaussianMixture(n_components=NUM_DIM, max_iter=100, covariance_type='full',
random_state=SEED)
gmm.fit(X_train)
X_train_gmm = gmm._estimate_weighted_log_prob(X_train)
X_score_gmm = gmm._estimate_weighted_log_prob(X_score)
# ====== rbm ====== #
rbm = BernoulliRBM(n_components=NUM_DIM, batch_size=8, learning_rate=0.0008,
n_iter=8, verbose=2, random_state=SEED)
rbm.fit(X_train)
X_train_rbm = rbm.transform(X_train)
X_score_rbm = rbm.transform(X_score)
# ===========================================================================
X_backend = lda.fit_transform(X=X_backend, y=y_backend)
lda_transform = lda.transform
else:
lda_transform = lambda x: x
# ====== training the PLDA ====== #
plda = PLDA(n_phi=N_PLDA,
centering=True,
wccn=True,
unit_length=True,
n_iter=20,
random_state=Config.SUPER_SEED,
verbose=2 if PLDA_SHOW_LLK else 1)
if PLDA_MAXIMUM_LIKELIHOOD:
print(" Fitting PLDA maximum likelihood ...")
plda.fit_maximum_likelihood(X=lda_transform(X_backend), y=y_backend)
plda.fit(X=lda_transform(X_backend), y=y_backend)
# ===========================================================================
# Now scoring
# ===========================================================================
for dsname, scores in sorted(all_vectors.items(), key=lambda x: x[0]):
# ====== skip non scoring dataset ====== #
if dsname not in SCORING_DATASETS:
continue
# ====== proceed ====== #
print("Scoring:", ctext(dsname, 'yellow'))
# load the scores
(seg_name, seg_meta, seg_path, seg_data) = (scores['name'], scores['y'],
scores['path'], scores['X'])
name_2_data = {i: j for i, j in zip(seg_name, seg_data)}
name_2_ext = {
i: '' if j is None else os.path.splitext(j)[-1]