path_model = '/home/aneumann/Schreibtisch/Net Models/my_model_db_smalllabels1.h5' path_data = '/media/aneumann/Harddisk/Datenbanken/PythonTest' path_data_save = '/media/aneumann/Harddisk/Datenbanken/PythonTest_2' path_save = '/media/aneumann/Harddisk/Datenbanken/PythonTest_5' skipcount = 2 framelength = 4 option = 'createdata' option2 = 'music' import ctypes import os import numpy as np import scipy.io import math from tqdm import tqdm os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' os.environ['TF_ENABLE_XLA'] = '1' os.environ['TF_ENABLE_AUTO_MIXED_PRECISION'] = '1' os.environ['TF_ENABLE_CUDNN_RNN_TENSOR_OP_MATH_FP32'] = '1' os.environ['TF_DISABLE_CUDNN_TENSOR_OP_MATH'] = '1' os.environ['TF_ENABLE_CUBLAS_TENSOR_OP_MATH_FP32'] = '1' from keras.models import load_model import tensorflow as tf gpus = tf.config.experimental.list_physical_devices('GPU') if gpus: try: tf.config.experimental.set_virtual_device_configuration(gpus[0], [tf.config.experimental.VirtualDeviceConfiguration(memory_limit=4096)]) except RuntimeError as e: print(e) model = tf.keras.models.load_model(path_model) # create data # ------------------------------------------------------------------------------------------------- if option == 'createdata': print('Creating files: 1weiter ...') from Preprocessing_flatten_1weiter import prep_flatten_1weiter diffcount = prep_flatten_1weiter(path_data, path_data_save, skipcount, framelength, option2) files_all = os.listdir(path_data_save) files_all = sorted(files_all) path_load = path_data_save elif option == 'getdata': print('Getting files ...') files_all = os.listdir(path_data) files_all = sorted(files_all) path_load = path_data else: print('Option nicht zulaessig!') # path = '/media/aneumann/Harddisk/Datenbanken/PythonTest_2/Processed_0_2_flatten_1weiter.npz' # data = np.load(path) # # x, y = data['x'], data['y'] # print(x.shape) # print(y.shape) # get data # ------------------------------------------------------------------------------------------------- print('Predicting ...') for file in files_all: path_loadcontents = path_load + '/' + file contents = np.load(path_loadcontents) X_raw, Y_raw = contents['x'], contents['y'] X_predicted_matrix = [] predictionError_matrix = [] dbmixed_matrix = [] dblabel_matrix = [] for m in tqdm(list(range(0,int(X_raw.shape[0]/diffcount)))): X, Y = X_raw[m*diffcount:m*diffcount+diffcount,:], Y_raw[m*diffcount:m*diffcount+diffcount,:] dbmixed = [] dblabel = [] X_predicted_whole = [] predictionError_whole = [] for n in list(range(0,diffcount)): X_predict, Y_predict = X[n,:], Y[n,:] X_predict, Y_predict = np.reshape(X_predict,(1,X_predict.shape[0])), np.reshape(Y_predict,(1,Y_predict.shape[0])) dbmixed.extend(X_predict[:,-(int(X_predict.shape[1]/framelength)):]) dblabel.extend(Y_predict) X_predicted = model.predict(X_predict) X_predicted_whole.extend(X_predicted) predictionError = Y_predict - X_predicted predictionError_whole.extend(predictionError) X_predicted_whole = np.transpose(X_predicted_whole) predictionError_whole = np.transpose(predictionError_whole) dbmixed = np.transpose(dbmixed) dblabel = np.transpose(dblabel) if m == 0: X_predicted_matrix = np.expand_dims(X_predicted_whole, axis = 0) predictionError_matrix = np.expand_dims(predictionError_whole, axis = 0) dbmixed_matrix = np.expand_dims(dbmixed, axis = 0) dblabel_matrix = np.expand_dims(dblabel, axis = 0) else: X_predicted_matrix = np.concatenate((X_predicted_matrix, np.expand_dims(X_predicted_whole, axis = 0)), axis=0) predictionError_matrix = np.concatenate((predictionError_matrix, np.expand_dims(predictionError_whole, axis = 0)), axis=0) dbmixed_matrix = np.concatenate((dbmixed_matrix, np.expand_dims(dbmixed, axis = 0)), axis=0) dblabel_matrix = np.concatenate((dblabel_matrix, np.expand_dims(dblabel, axis = 0)), axis=0) # saving # ------------------------------------------------------------------------------------------------- index = file.split("_",2) path_savecontents = path_save + '/Predicted_' + index[1] + '_' + str(X_predicted_matrix.shape[0]) + '_Dense_1weiter.npz' np.savez(path_savecontents, X_predicted_matrix=X_predicted_matrix, predictionError_matrix=predictionError, dbmixed_matrix=dbmixed_matrix, dblabel_matrix=dblabel_matrix) print(X_predicted_matrix.shape)