def plot_latent(self, cur_epoch=''):
# Generating latent space
print('Generating latent space ...')
latent_en = self.encode(self.data_plot.x)
pca = PCA(n_components=2)
latent_pca = latent_en if latent_en.shape[1] == 2 else latent_en[:, 0:2] if latent_en.shape[1]==3 else pca.fit_transform(latent_en)
print('Latent space dimensions: {}'.format(latent_pca.shape))
print('Plotting latent space ...')
latent_space = self.config.log_dir + '/latent2d/{} latent epoch {}.jpg'.format(self.config.log_dir.split('/')[-1:][0],
cur_epoch)
self.latent_space_files.append(latent_space)
plot_dataset(latent_pca.compute(), y=self.data_plot.labels, save=latent_space)
if latent_en.shape[1] >= 3:
pca = PCA(n_components=3)
latent_pca = latent_en if latent_en.shape[1]==3 else pca.fit_transform(latent_en)
print('latent space dimensions: {}'.format(latent_pca.shape))
print('Plotting latent space ...')
latent_space = self.config.log_dir + '/latent3d/{} latent_3d epoch {}.jpg'.format(self.config.log_dir.split('/')[-1:][0],
cur_epoch)
self.latent_space3d_files.append(latent_space)
plot_dataset3d(latent_pca.compute(), y=self.data_plot.labels, save=latent_space)
del latent_pca, latent_en
gc.collect()
def do_pca(g, n_comp):
"""
Perform a PCA on the genetic array and return n_comp of it
:param g: Genotype array
:param n_comp: Number of components sought
:return: components array
"""
pca = PCA(n_components=n_comp)
pca = pca.fit_transform(g)
return pca
def do_pca(g, n_comp):
"""
Perform a PCA on the genetic array and return n_comp of it
:param g: Genotype array
:param n_comp: Number of components sought
:return: components array
"""
cache = Chest(available_memory=available_memory, path=os.getcwd())
pca = PCA(n_components=n_comp)
pca = pca.fit_transform(g)
return pca.compute(cache=cache)
def generate_samples(self, data, session, cur_epoch=''):
# Generating W space
print('Generating W space ...')
w_en = self.encode(data.x)
pca = PCA(n_components=2)
W_pca = pca.fit_transform(w_en)
print('W space dimensions: {}'.format(W_pca.shape))
print('Ploting W space ...')
w_space = self.summary_dir + '/{} W space in epoch {}.jpg'.format(
self.summary_dir.split('/')[-1:][0], cur_epoch)
self.w_space_files.append(w_space)
plot_dataset(W_pca.compute(), y=data.labels, save=w_space)
pca = PCA(n_components=3)
W_pca = pca.fit_transform(w_en)
print('W space dimensions: {}'.format(W_pca.shape))
print('Ploting W space ...')
w_space = self.summary_dir + '/{} W space 3d in epoch {}.jpg'.format(
self.summary_dir.split('/')[-1:][0], cur_epoch)
self.w_space3d_files.append(w_space)
plot_dataset3d(W_pca.compute(), y=data.labels, save=w_space)
del W_pca, w_en
gc.collect()
# Generating Samples
print('Generating Samples ...')
x_recons_l = self.reconst(data.samples)
recons_file = self.summary_dir + '/{} samples generation in epoch {}.jpg'.format(
self.summary_dir.split('/')[-1:][0], cur_epoch)
self.recons_files.append(recons_file)
plot_samples(x_recons_l, scale=10, save=recons_file)
del x_recons_l
gc.collect()
def plot_latent(self, cur_epoch=''):
# Generating latent space
if self.latent_data is None:
self.generate_latent(self.data_train, self.session,
self.config.ntrain_batches)
pca = PCA(n_components=2)
latent_pca = self.latent_data['latent'] if self.latent_data['latent'].shape[1] == 2 \
else self.latent_data['latent'][:, 0:2] if self.latent_data['latent'].shape[1]==3 \
else pca.fit_transform(self.latent_data['latent'])
print('Latent space dimensions: {}'.format(latent_pca.shape))
print('Plotting latent space ...')
latent_space = self.config.log_dir + '/latent2d/{} latent epoch {}.jpg'.format(
self.config.log_dir.split('/')[-1:][0], cur_epoch)
plot_dataset(
latent_pca.compute(),
y=self.latent_data['label'][:,
self.latent_data['y_index']].compute(),
save=latent_space)
if self.latent_data['latent'].shape[1] >= 3:
pca = PCA(n_components=3)
latent_pca = self.latent_data['latent'] if self.latent_data['latent'].shape[1]==3 \
else pca.fit_transform(self.latent_data['latent'])
print('latent space dimensions: {}'.format(latent_pca.shape))
print('Plotting latent space ...')
latent_space = self.config.log_dir + '/latent3d/{} latent_3d epoch {}.jpg'.format(
self.config.log_dir.split('/')[-1:][0], cur_epoch)
plot_dataset3d(latent_pca.compute(),
y=self.latent_data['label']
[:, self.latent_data['y_index']].compute(),
save=latent_space)
del latent_pca
gc.collect()
def pca(dataset: DataSet, num_components: int):
pca = PCA(n_components=num_components)
reduced = pca.fit_transform(dataset.feature_vector)
components = pca.components_
return DataSet(reduced), components