metavar='path',
type=str,
help='the path to save the image results ')
args = my_parser.parse_args()
image_directory = args.image_directory
feature_directory = args.feature_directory
results_directory = args.results_directory
files_path = image_utils.loadFilePaths(image_directory)
for file in list(os.listdir(feature_directory)):
if file.endswith('.npy'):
print(f'{feature_directory}/{file}')
features = image_utils.loadFeatures(
os.path.join(feature_directory, file))
print("tsne....")
tsne_transformed = TSNE(n_components=3,
n_jobs=-1).fit_transform(features)
output_df = runHierarchicalClustering(
tsne_transformed,
f'{results_directory}/{file}-hierarchical-tsne')
output_df.to_csv(
f'{results_directory}/{file}-hierarchical-tsne.csv')
print("pca....")
pca_dims = PCA().fit(features)
cumsum = np.cumsum(pca_dims.explained_variance_ratio_)
d = np.argmax(cumsum >= 0.95) + 1
print(d)
if (d == 1):
if __name__ == "__main__":
init = 'glorot_uniform'
pretrain_optimizer = 'adam'
# setting parameters
update_interval = 140
pretrain_epochs = 300
#pretrain_epochs=1
init = VarianceScaling(scale=1. / 3., mode='fan_in',
distribution='uniform') # [-limit, limit], limit=sqrt(1./fan_in)
pretrain_optimizer = SGD(lr=1, momentum=0.9)
#x, y = load_mnist()
#n_clusters = len(np.unique(y))
features=image_utils.loadFeatures('image_features_autoencoder_helper/ideology__grayscale_.npy')
data=features
data = data.astype('float32') / 255.
x=data
y=None
print(x.shape)
n_clusters=500
# load the gray scale features in the variable x define y to None
dec = DEC(dims=[x.shape[-1], 500, 500, 2000, 10], n_clusters=n_clusters, init=init)
dec.pretrain(x=x, y=y, optimizer=pretrain_optimizer,epochs=pretrain_epochs, batch_size=256,save_dir='')
print("DONE pretraining")
dec.compile(optimizer=SGD(0.01, 0.9), loss='kld')
y_pred = dec.fit(x, y=y, tol=0.001, maxiter=2e4, batch_size=256,update_interval=update_interval, save_dir='')
np.save("dec_predictions.npy",y_pred)
print("Saved")
#saving the y_pred labels and making the pdf out of it