def plot_sample_angle_combined(train_features,
train_labels,
test_features,
test_labels,
model_dir,
title1,
title2,
tail=""):
save_dir = os.path.join(model_dir, "figures", "sample_angle_combined")
os.makedirs(save_dir, exist_ok=True)
colors = ['blue', 'red', 'green']
_bins = np.linspace(-0.05, 1.05, 21)
classes = np.unique(y_train)
fs_train, _ = utils.sort_dataset(train_features,
train_labels,
classes=classes,
stack=False)
fs_test, _ = utils.sort_dataset(test_features,
test_labels,
classes=classes,
stack=False)
angles = []
for class_train in classes:
for class_test in classes:
if class_train == class_test:
continue
angles.append(
(fs_train[class_train] @ fs_test[class_test].T).reshape(-1))
# plt.rc('text', usetex=True)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman']
fig, ax = plt.subplots(figsize=(7, 5))
ax.hist(
np.hstack(angles),
bins=_bins,
alpha=0.5,
color='red', #colors[class_test],
edgecolor='black') #, label=f'Class {class_test}')
ax.set_xlabel('Similarity', fontsize=38)
ax.set_ylabel('Count', fontsize=38)
ax.ticklabel_format(style='sci', scilimits=(0, 3))
[tick.label.set_fontsize(22) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(22) for tick in ax.yaxis.get_major_ticks()]
# ax.legend(loc='upper center', prop={"size": 13}, ncol=1, framealpha=0.5)
fig.tight_layout()
fig.savefig(
os.path.join(save_dir,
f'sample_angle_combined-{title1}-vs-{title2}{tail}.pdf'))
plt.close()
def plot_nearest_component_class(args, features, labels, epoch, trainset):
"""Find corresponding images to the nearests component per class. """
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
data_sort, _ = utils.sort_dataset(trainset.data,
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
for class_ in range(trainset.num_classes):
nearest_data = []
nearest_val = []
pca = TruncatedSVD(n_components=10,
random_state=10).fit(features_sort[class_])
for j in range(8):
proj = features_sort[class_] @ pca.components_.T[:, j]
img_idx = np.argsort(np.abs(proj), axis=0)[::-1][:10]
nearest_val.append(proj[img_idx])
nearest_data.append(np.array(data_sort[class_])[img_idx])
fig, ax = plt.subplots(ncols=10, nrows=8, figsize=(10, 10))
for r in range(8):
for c in range(10):
ax[r, c].imshow(nearest_data[r][c])
ax[r, c].set_xticks([])
ax[r, c].set_yticks([])
ax[r, c].spines['top'].set_visible(False)
ax[r, c].spines['right'].set_visible(False)
ax[r, c].spines['bottom'].set_linewidth(False)
ax[r, c].spines['left'].set_linewidth(False)
ax[r, c].set_xlabel(f"proj: {nearest_val[r][c]:.2f}")
if c == 0:
ax[r, c].set_ylabel(f"comp {r}")
fig.tight_layout()
## save
save_dir = os.path.join(args.model_dir, 'figures', 'nearcomp_class')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"nearest_class{class_}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"nearest_class{class_}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def nearsub(train_features,
train_labels,
test_features,
test_labels,
num_classes,
n_comp=10,
return_pred=False):
train_scores, test_scores = [], []
classes = np.arange(num_classes)
features_sort, _ = utils.sort_dataset(train_features,
train_labels,
classes=classes,
stack=False)
fd = features_sort[0].shape[1]
for j in classes:
_, _, V = torch.svd(features_sort[j])
components = V[:, :n_comp].T
subspace_j = torch.eye(fd) - components.T @ components
train_j = subspace_j @ train_features.T
test_j = subspace_j @ test_features.T
train_scores_j = torch.linalg.norm(train_j, ord=2, axis=0)
test_scores_j = torch.linalg.norm(test_j, ord=2, axis=0)
train_scores.append(train_scores_j)
test_scores.append(test_scores_j)
train_pred = torch.stack(train_scores).argmin(0)
test_pred = torch.stack(test_scores).argmin(0)
if return_pred:
return train_pred.numpy(), test_pred.numpy()
train_acc = compute_accuracy(classes[train_pred], train_labels.numpy())
test_acc = compute_accuracy(classes[test_pred], test_labels.numpy())
print('SVD: {}, {}'.format(train_acc, test_acc))
return train_acc, test_acc
def plot_heatmap(model_dir, name, features, labels, num_classes):
"""Plot heatmap of cosine simliarity for all features. """
features_sort, _ = utils.sort_dataset(features,
labels,
classes=num_classes,
stack=False)
features_sort_ = np.vstack(features_sort)
sim_mat = np.abs(features_sort_ @ features_sort_.T)
# plt.rc('text', usetex=False)
# plt.rcParams['font.family'] = 'serif'
# plt.rcParams['font.serif'] = ['Times New Roman'] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(figsize=(7, 5), sharey=True, sharex=True)
im = ax.imshow(sim_mat, cmap='Blues')
fig.colorbar(im, pad=0.02, drawedges=0, ticks=[0, 0.5, 1])
ax.set_xticks(np.linspace(0, len(labels), num_classes + 1))
ax.set_yticks(np.linspace(0, len(labels), num_classes + 1))
[tick.label.set_fontsize(10) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(10) for tick in ax.yaxis.get_major_ticks()]
fig.tight_layout()
save_dir = os.path.join(model_dir, 'figures', 'heatmaps')
os.makedirs(save_dir, exist_ok=True)
file_name = os.path.join(save_dir, f"{name}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def nearsub(args, train_features, train_labels, test_features, test_labels):
"""Perform nearest subspace classification.
Options:
n_comp (int): number of components for PCA or SVD
"""
scores_pca = []
scores_svd = []
num_classes = train_labels.numpy().max() + 1 # should be correct most of the time
features_sort, _ = utils.sort_dataset(train_features.numpy(), train_labels.numpy(),
num_classes=num_classes, stack=False)
for j in range(num_classes):
pca = PCA(n_components=args.n_comp).fit(features_sort[j])
pca_subspace = pca.components_.T
mean = np.mean(features_sort[j], axis=0)
pca_j = (np.eye(params["fd"]) - pca_subspace @ pca_subspace.T) \
@ (test_features.numpy() - mean).T
score_pca_j = np.linalg.norm(pca_j, ord=2, axis=0)
svd = TruncatedSVD(n_components=args.n_comp).fit(features_sort[j])
svd_subspace = svd.components_.T
svd_j = (np.eye(params["fd"]) - svd_subspace @ svd_subspace.T) \
@ (test_features.numpy()).T
score_svd_j = np.linalg.norm(svd_j, ord=2, axis=0)
scores_pca.append(score_pca_j)
scores_svd.append(score_svd_j)
test_predict_pca = np.argmin(scores_pca, axis=0)
test_predict_svd = np.argmin(scores_svd, axis=0)
acc_pca = utils.compute_accuracy(test_predict_pca, test_labels.numpy())
acc_svd = utils.compute_accuracy(test_predict_svd, test_labels.numpy())
print('PCA: {}'.format(acc_pca))
print('SVD: {}'.format(acc_svd))
return acc_pca
def plot_heatmap(features, labels, title, model_dir):
"""Plot heatmap of cosine simliarity for all features. """
num_samples = features.shape[0]
classes = np.arange(np.unique(labels).size)
features_sort_, _ = utils.sort_dataset(features,
labels,
classes=classes,
stack=True)
sim_mat = np.abs(features_sort_ @ features_sort_.T)
print(sim_mat.min(), sim_mat.max())
# plt.rc('text', usetex=True)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman']
fig, ax = plt.subplots(figsize=(8, 7), sharey=True, sharex=True)
im = ax.imshow(sim_mat, cmap='Blues')
divider = make_axes_locatable(ax)
cax = divider.append_axes("right", size="5%", pad=0.1)
cbar = fig.colorbar(im, cax=cax, drawedges=0, ticks=[0, 0.5, 1])
cbar.ax.tick_params(labelsize=18)
# fig.colorbar(im, pad=0.02, drawedges=0, ticks=[0, 0.5, 1])
ax.set_xticks(np.linspace(0, num_samples, len(classes) + 1))
ax.set_yticks(np.linspace(0, num_samples, len(classes) + 1))
[tick.label.set_fontsize(24) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(24) for tick in ax.yaxis.get_major_ticks()]
fig.tight_layout()
save_dir = os.path.join(model_dir, "figures", "heatmaps")
os.makedirs(save_dir, exist_ok=True)
plt.savefig(os.path.join(save_dir, f"heatmap-{title}.pdf"))
plt.close()
def nearsub(train_features,
train_labels,
test_features,
test_labels,
n_comp=10):
"""Perform nearest subspace classification.
Options:
n_comp (int): number of components for PCA or SVD
"""
scores_svd = []
classes = np.unique(test_labels)
features_sort, _ = utils.sort_dataset(train_features,
train_labels,
classes=classes,
stack=False)
fd = features_sort[0].shape[1]
if n_comp >= fd:
n_comp = fd - 1
for j in np.arange(len(classes)):
svd = TruncatedSVD(n_components=n_comp).fit(features_sort[j])
svd_subspace = svd.components_.T
svd_j = (np.eye(fd) - svd_subspace @ svd_subspace.T) \
@ (test_features).T
score_svd_j = np.linalg.norm(svd_j, ord=2, axis=0)
scores_svd.append(score_svd_j)
test_predict_svd = np.argmin(scores_svd, axis=0)
acc_svd = compute_accuracy(classes[test_predict_svd], test_labels)
print('SVD: {}'.format(acc_svd))
return acc_svd
def nearsub_pca(train_features,
train_labels,
test_features,
test_labels,
n_comp=10):
"""Perform nearest subspace classification.
Options:
n_comp (int): number of components for PCA or SVD
"""
scores_pca = []
classes = np.unique(test_labels)
features_sort, _ = utils.sort_dataset(train_features,
train_labels,
classes=classes,
stack=False)
fd = features_sort[0].shape[1]
if n_comp >= fd:
n_comp = fd - 1
for j in np.arange(len(classes)):
pca = PCA(n_components=n_comp).fit(features_sort[j])
pca_subspace = pca.components_.T
mean = np.mean(features_sort[j], axis=0)
pca_j = (np.eye(fd) - pca_subspace @ pca_subspace.T) \
@ (test_features - mean).T
score_pca_j = np.linalg.norm(pca_j, ord=2, axis=0)
scores_pca.append(score_pca_j)
test_predict_pca = np.argmin(scores_pca, axis=0)
acc_pca = compute_accuracy(classes[test_predict_pca], test_labels)
print('PCA: {}'.format(acc_pca))
return acc_svd
def get_dataset(self):
dataset = open(self.file_path,encoding='utf-8').readlines()
new_dataset = []
sent = []
label_counts = Counter()
root = [START_TAG, START_TAG]
for line in dataset:
if len(line.strip())==0:
if len(sent)>0:
sent.append([END_TAG, END_TAG ])
if len(sent)>2:
new_dataset.append([root]+sent)
#new_dataset.append(sent)
sent = []
else:
row = line.rstrip().split()
row[0] = row[0].replace("\ufeff","")
sent.append(row)
label_counts.update([row[-1]])
if len(sent)>0:
sent.append([END_TAG, END_TAG ])
new_dataset.append([root]+sent)
#new_dataset.append(sent)
print("Number of sentences : {} ".format(len(new_dataset)))
#print(new_dataset)
new_dataset, orig_idx = sort_dataset(new_dataset, sort = True)
return new_dataset, orig_idx, label_counts
def plot_hist(args, features, labels, epoch):
"""Plot histogram of class vs. class. """
## create save folder
hist_folder = os.path.join(args.model_dir, 'figures', 'hist')
if not os.path.exists(hist_folder):
os.makedirs(hist_folder)
num_classes = labels.numpy().max() + 1
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=num_classes,
stack=False)
for i in range(num_classes):
for j in range(i, num_classes):
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(7, 5), dpi=250)
if i == j:
sim_mat = features_sort[i] @ features_sort[j].T
sim_mat = sim_mat[np.triu_indices(sim_mat.shape[0], k=1)]
else:
sim_mat = (features_sort[i] @ features_sort[j].T).reshape(-1)
ax.hist(sim_mat, bins=40, color='red', alpha=0.5)
ax.set_xlabel("cosine similarity")
ax.set_ylabel("count")
ax.set_title(f"Class {i} vs. Class {j}")
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
fig.tight_layout()
file_name = os.path.join(hist_folder, f"hist_{i}v{j}")
fig.savefig(file_name)
plt.close()
print("Plot saved to: {}".format(file_name))
def plot_heatmap(args, features, labels, epoch):
"""Plot heatmap of cosine simliarity for all features. """
num_classes = trainset.num_classes
features_sort, _ = utils.sort_dataset(features.numpy(), labels.numpy(),
num_classes=num_classes, stack=False)
features_sort_ = np.vstack(features_sort)
sim_mat = np.abs(features_sort_ @ features_sort_.T)
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(figsize=(7, 5), sharey=True, sharex=True, dpi=400)
im = ax.imshow(sim_mat, cmap='Blues')
fig.colorbar(im, pad=0.02, drawedges=0, ticks=[0, 0.5, 1])
ax.set_xticks(np.linspace(0, 50000, 6))
ax.set_yticks(np.linspace(0, 50000, 6))
[tick.label.set_fontsize(10) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(10) for tick in ax.yaxis.get_major_ticks()]
fig.tight_layout()
save_dir = os.path.join(args.model_dir, 'figures', 'heatmaps')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"heatmat_epoch{epoch}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"heatmat_epoch{epoch}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_pca(args, features, labels, epoch):
"""Plot PCA of learned features. """
## create save folder
pca_dir = os.path.join(args.model_dir, 'figures', 'pca')
if not os.path.exists(pca_dir):
os.makedirs(pca_dir)
## perform PCA on features
n_comp = np.min([args.comp, features.shape[1]])
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
pca = PCA(n_components=n_comp).fit(features.numpy())
sig_vals = [pca.singular_values_]
for c in range(trainset.num_classes):
pca = PCA(n_components=n_comp).fit(features_sort[c])
sig_vals.append((pca.singular_values_))
## plot features
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(7, 5), dpi=500)
x_min = np.min([len(sig_val) for sig_val in sig_vals])
ax.plot(np.arange(x_min),
sig_vals[0][:x_min],
'-p',
markersize=3,
markeredgecolor='black',
linewidth=1.5,
color='tomato')
map_vir = plt.cm.get_cmap('Blues', 6)
norm = plt.Normalize(-10, 10)
class_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
norm_class = norm(class_list)
color = map_vir(norm_class)
for c, sig_val in enumerate(sig_vals[1:]):
ax.plot(np.arange(x_min),
sig_val[:x_min],
'-o',
markersize=3,
markeredgecolor='black',
alpha=0.6,
linewidth=1.0,
color=color[c])
ax.set_xticks(np.arange(0, x_min, 5))
ax.set_yticks(np.arange(0, 35, 5))
ax.set_xlabel("components", fontsize=14)
ax.set_ylabel("sigular values", fontsize=14)
[tick.label.set_fontsize(12) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(12) for tick in ax.yaxis.get_major_ticks()]
fig.tight_layout()
np.save(os.path.join(pca_dir, "sig_vals.npy"), sig_vals)
file_name = os.path.join(pca_dir, f"pca_classVclass_epoch{epoch}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(pca_dir, f"pca_classVclass_epoch{epoch}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_nearest_component_supervised(args, features, labels, epoch, trainset):
"""Find corresponding images to the nearests component. """
## perform PCA on features
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
data_sort, _ = utils.sort_dataset(trainset.data,
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
nearest_data = []
for c in range(trainset.num_classes):
pca = TruncatedSVD(n_components=10,
random_state=10).fit(features_sort[c])
proj = features_sort[c] @ pca.components_.T
img_idx = np.argmax(np.abs(proj), axis=0)
nearest_data.append(np.array(data_sort[c])[img_idx])
fig, ax = plt.subplots(ncols=10, nrows=10, figsize=(10, 10))
for r in range(10):
for c in range(10):
ax[r, c].imshow(nearest_data[r][c])
ax[r, c].set_xticks([])
ax[r, c].set_yticks([])
ax[r, c].spines['top'].set_visible(False)
ax[r, c].spines['right'].set_visible(False)
ax[r, c].spines['bottom'].set_linewidth(False)
ax[r, c].spines['left'].set_linewidth(False)
if c == 0:
ax[r, c].set_ylabel(f"comp {r}")
## save
save_dir = os.path.join(args.model_dir, 'figures', 'nearcomp_sup')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"nearest_data.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"nearest_data.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_nearest_component(args, features, labels, epoch, trainset):
## perform PCA on features
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=len(trainset.classes),
stack=False)
data_sort, _ = utils.sort_dataset(trainset.data,
labels.numpy(),
num_classes=len(trainset.classes),
stack=False)
nearest_data = []
for c in range(len(trainset.classes)):
pca = TruncatedSVD(n_components=10,
random_state=10).fit(features_sort[c])
proj = features_sort[c] @ pca.components_.T
img_idx = np.argmax(np.abs(proj), axis=0)
nearest_data.append(np.array(data_sort[c])[img_idx])
fig, ax = plt.subplots(ncols=10, nrows=10, figsize=(10, 10))
for r in range(10):
for c in range(10):
ax[r, c].imshow(nearest_data[r][c])
ax[r, c].set_axis_off()
plt.xlabel("per component")
plt.ylabel("per class")
## save
save_dir = os.path.join(args.model_dir, 'figures', 'pca')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"nearest_data.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"nearest_data.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_heatmap(args, features, labels, epoch):
num_classes = len(trainset.classes)
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=num_classes,
stack=False)
features_sort_ = np.vstack(features_sort)
sim_mat = np.abs(features_sort_ @ features_sort_.T)
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'
] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(figsize=(7, 5), sharey=True, sharex=True, dpi=400)
im = ax.imshow(sim_mat, cmap='Blues')
fig.colorbar(im, pad=0.02, drawedges=0, ticks=[0, 0.5, 1])
ax.set_xticks(np.linspace(0, 50000, 6))
ax.set_yticks(np.linspace(0, 50000, 6))
# ax.spines['right'].set_visible(False)
# ax.spines['top'].set_visible(False)
# ax.spines['bottom'].set_visible(False)
# ax.spines['left'].set_visible(False)
[tick.label.set_fontsize(10) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(10) for tick in ax.yaxis.get_major_ticks()]
fig.tight_layout()
save_dir = os.path.join(args.model_dir, 'figures', 'heatmaps')
# os.mkdir(save_dir+"/sim_mat/")
for i in range(500):
np.save(save_dir + f"/sim_mat/sim_mat{i}.npy",
sim_mat[i * 100:(i + 1) * 100])
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"heatmat_epoch{epoch}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"heatmat_epoch{epoch}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_pca(args, features, labels, epoch):
## create save folder
pca_dir = os.path.join(args.model_dir, 'figures', 'pca')
if not os.path.exists(pca_dir):
os.makedirs(pca_dir)
## perform PCA on features
n_comp = np.min([args.comp, features.shape[1]])
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=len(trainset.classes),
stack=False)
pca = PCA(n_components=n_comp).fit(features.numpy())
sig_vals = [pca.singular_values_]
for c in range(len(trainset.classes)):
pca = PCA(n_components=n_comp).fit(features_sort[c])
sig_vals.append((pca.singular_values_))
## plot features
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'
] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(7, 5), dpi=500)
x_min = np.min([len(sig_val) for sig_val in sig_vals])
ax.plot(np.arange(x_min),
sig_vals[0][:x_min],
'-p',
markersize=3,
markeredgecolor='black',
linewidth=1.5,
color='tomato')
map_vir = plt.cm.get_cmap('Blues', 6)
norm = plt.Normalize(-10, 10)
class_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
norm_class = norm(class_list)
color = map_vir(norm_class)
for c, sig_val in enumerate(sig_vals[1:]):
ax.plot(np.arange(x_min),
sig_val[:x_min],
'-o',
markersize=3,
markeredgecolor='black',
alpha=0.6,
linewidth=1.0,
color=color[c])
ax.set_xticks(np.arange(0, x_min, 5))
ax.set_yticks(np.arange(0, 35, 5))
# for c, sig_val in enumerate(sig_vals[1:]):
# ax.plot(np.arange(x_min), sig_val[:x_min], marker='o', markersize=2,
# label=f'class - {c}', alpha=0.6)
# ax.legend(loc='upper right', frameon=True, fancybox=True, prop={"size": 12}, ncol=2, framealpha=0.5)
ax.set_xlabel("components", fontsize=14)
ax.set_ylabel("sigular values", fontsize=14)
# ax.set_title(f"PCA on features (Epoch: {epoch})")
# ax.spines['top'].set_visible(False)
# ax.spines['right'].set_visible(False)
# ax.spines['bottom'].set_visible(False)
# ax.spines['left'].set_visible(False)
[tick.label.set_fontsize(12) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(12) for tick in ax.yaxis.get_major_ticks()]
# ax.grid(True, color='white')
# ax.set_facecolor('whitesmoke')
fig.tight_layout()
np.save(os.path.join(pca_dir, "sig_vals.npy"), sig_vals)
file_name = os.path.join(pca_dir, f"pca_classVclass_epoch{epoch}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(pca_dir, f"pca_classVclass_epoch{epoch}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_hist_all(args, features, labels, epoch):
## create save folder
hist_folder = os.path.join(args.model_dir, 'figures', 'hist_all')
if not os.path.exists(hist_folder):
os.makedirs(hist_folder)
if not os.path.exists(os.path.join(hist_folder, "sim_mat")):
os.makedirs(os.path.join(hist_folder, "sim_mat"))
num_classes = len(trainset.classes)
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=num_classes,
stack=False)
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman']
fig, ax = plt.subplots(ncols=10,
nrows=10,
figsize=(10, 10),
dpi=250,
sharex=True,
sharey=True)
for j in range(10):
for i in range(10):
sim_mat = features_sort[i] @ features_sort[j].T
if i == j:
sim_mat = sim_mat[np.triu_indices(sim_mat.shape[0], k=1)]
h1 = ax[i, j].hist(sim_mat,
bins=np.arange(10) - 0.05,
color='green',
alpha=0.3,
histtype='bar',
density=True)
else:
sim_mat = sim_mat.reshape(-1)
h2 = ax[i, j].hist(sim_mat,
bins=np.arange(10) - 0.05,
color='red',
alpha=0.3,
histtype='bar',
density=True)
ax[i, j].spines['right'].set_visible(False)
ax[i, j].spines['top'].set_visible(False)
ax[i, j].spines['bottom'].set_visible(False)
ax[i, j].spines['left'].set_visible(False)
ax[i, j].grid(True, color='white', axis='y')
ax[i, j].set_xlim(0, 1.)
ax[i, j].set_facecolor('whitesmoke')
if i == 0:
ax[i, j].set_title(f'class {j}')
if j == 0:
ax[i, j].set_ylabel(f'class {i}')
np.save(
os.path.join(hist_folder, "sim_mat", f"sim_mat{i}_{j}.npy"),
sim_mat)
fig.text(0.5, -0.01, 'cosine similarity', ha='center')
fig.text(-0.01, 0.5, 'count', va='center', rotation='vertical')
plt.tight_layout()
plt.show()
file_name = os.path.join(hist_folder, f"hist_all_epoch{epoch}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(hist_folder, f"hist_all_epoch{epoch}.pdf")
fig.savefig(file_name)
plt.close()
print("Plot saved to: {}".format(file_name))
def plot_pca_epoch(args):
"""Plot PCA for different epochs in the same plot. """
EPOCHS = [0, 10, 100, 500]
params = utils.load_params(args.model_dir)
transforms = tf.load_transforms('test')
trainset = tf.load_trainset(params['data'], transforms)
trainloader = DataLoader(trainset, batch_size=200, num_workers=4)
sig_vals = []
for epoch in EPOCHS:
epoch_ = epoch - 1
if epoch_ == -1: # randomly initialized
net = tf.load_architectures(params['arch'], params['fd'])
else:
net, epoch = tf.load_checkpoint(args.model_dir,
epoch=epoch_,
eval_=True)
features, labels = tf.get_features(net, trainloader)
if args.class_ is not None:
features_sort, _ = utils.sort_dataset(
features.numpy(),
labels.numpy(),
num_classes=trainset.num_classes,
stack=False)
features_ = features_sort[args.class_]
else:
features_ = features.numpy()
n_comp = np.min([args.comp, features.shape[1]])
pca = PCA(n_components=n_comp).fit(features_)
sig_vals.append(pca.singular_values_)
## plot singular values
plt.rc('text', usetex=True)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman']
fig, ax = plt.subplots(1, 1, figsize=(7, 5), dpi=400)
x_min = np.min([len(sig_val) for sig_val in sig_vals])
if args.class_ is not None:
ax.set_xticks(np.arange(0, x_min, 10))
ax.set_yticks(np.linspace(0, 40, 9))
ax.set_ylim(0, 40)
else:
ax.set_xticks(np.arange(0, x_min, 10))
ax.set_yticks(np.linspace(0, 80, 9))
ax.set_ylim(0, 90)
for epoch, sig_val in zip(EPOCHS, sig_vals):
ax.plot(np.arange(x_min),
sig_val[:x_min],
marker='',
markersize=5,
label=f'epoch - {epoch}',
alpha=0.6)
ax.legend(loc='upper right',
frameon=True,
fancybox=True,
prop={"size": 8},
ncol=1,
framealpha=0.5)
ax.set_xlabel("components")
ax.set_ylabel("sigular values")
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['bottom'].set_visible(False)
ax.spines['left'].set_visible(False)
[tick.label.set_fontsize(12) for tick in ax.xaxis.get_major_ticks()]
[tick.label.set_fontsize(12) for tick in ax.yaxis.get_major_ticks()]
ax.grid(True, color='white')
ax.set_facecolor('whitesmoke')
fig.tight_layout()
## save
save_dir = os.path.join(args.model_dir, 'figures', 'pca')
np.save(os.path.join(save_dir, "sig_vals_epoch.npy"), sig_vals)
if not os.path.exists(save_dir):
os.makedirs(save_dir)
file_name = os.path.join(save_dir, f"pca_class{args.class_}.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(save_dir, f"pca_class{args.class_}.pdf")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
plt.close()
def plot_hist_paper(args, features, labels, epoch):
## create save folder
hist_folder = os.path.join(args.model_dir, 'figures', 'hist_paper')
if not os.path.exists(hist_folder):
os.makedirs(hist_folder)
num_classes = len(trainset.classes)
features_sort, _ = utils.sort_dataset(features.numpy(),
labels.numpy(),
num_classes=num_classes,
stack=False)
i = 0
## inside class
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'
] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(7, 5), dpi=250)
sim_mat = features_sort[i] @ features_sort[i].T
sim_mat = sim_mat[np.triu_indices(sim_mat.shape[0], k=1)][-500:]
ax.hist(sim_mat, bins=30, color='green', alpha=0.4, density=True)
ax.grid(True, color='white', axis='y')
ax.set_xlim(0, 1.)
ax.set_facecolor('whitesmoke')
ax.set_xlabel("cosine similarity")
ax.set_ylabel("density")
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
ax.spines['bottom'].set_visible(False)
ax.spines['left'].set_visible(False)
fig.tight_layout()
file_name = os.path.join(hist_folder, f"hist_0v0.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(hist_folder, f"hist_0v0.pdf")
fig.savefig(file_name)
plt.close()
print("Plot saved to: {}".format(file_name))
## outside class
plt.rc('text', usetex=False)
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'
] #+ plt.rcParams['font.serif']
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(7, 5), dpi=250)
temp = []
for j in range(num_classes):
sim_mat = (features_sort[i] @ features_sort[j].T).reshape(-1)
temp.append(sim_mat)
temp = np.hstack(temp)
ax.hist(temp, bins=30, color='red', alpha=0.4, density=True)
ax.grid(True, color='white', axis='y')
ax.set_xlim(0, 1.)
ax.set_facecolor('whitesmoke')
ax.set_xlabel("cosine similarity")
ax.set_ylabel("density")
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
ax.spines['bottom'].set_visible(False)
ax.spines['left'].set_visible(False)
fig.tight_layout()
file_name = os.path.join(hist_folder, f"hist_0vall.png")
fig.savefig(file_name)
print("Plot saved to: {}".format(file_name))
file_name = os.path.join(hist_folder, f"hist_0vall.pdf")
fig.savefig(file_name)
plt.close()
print("Plot saved to: {}".format(file_name))
