Ahhhh nice, yes
The problem was, there has always been a zombie process. That is way I thought the termination did not work properly. What I forgot to do is to also make the QRunnable quit, which remained in the main thread as still running.
This is the solution:
import tensorflow as tf
from sys import exit, argv
from multiprocessing import Process, Queue
from PyQt5.QtWidgets import QPushButton, QApplication, QHBoxLayout, QWidget, QLabel
from PyQt5.QtCore import QRunnable, QObject, pyqtSignal, QThreadPool
class Window(QWidget):
def __init__(self):
QWidget.__init__(self)
self.__btn_run = QPushButton("Start")
self.__btn_stp = QPushButton("Stop")
self.__label = QLabel("Idle")
self.__runner = Runner()
self.__pool = QThreadPool.globalInstance()
self.__btn_run.clicked.connect(self.__run_net)
self.__btn_stp.clicked.connect(self.__stp_net)
self.__runner.signals.finished.connect(self.__on_finished)
self.__btn_stp.setDisabled(True)
self.setLayout(QHBoxLayout())
self.layout().addWidget(self.__btn_run)
self.layout().addWidget(self.__btn_stp)
self.layout().addWidget(self.__label)
def __run_net(self):
self.__btn_run.setDisabled(True)
self.__btn_stp.setEnabled(True)
self.__label.setText("Running")
self.__pool.start(self.__runner)
def __stp_net(self):
self.__runner.close()
# What to do here?
def __on_finished(self):
self.__btn_run.setEnabled(True)
self.__btn_stp.setDisabled(True)
self.__label.setText("Finished")
self.__runner = Runner()
class Runner(QRunnable):
def __init__(self):
QRunnable.__init__(self)
self.__queue = Queue()
self.__net = Mnist(self.__queue)
self.signals = RunnerSignal()
def run(self):
self.__net.start()
while True:
data = self.__queue.get()
if data == NetSignal.finished:
self.signals.finished.emit()
break
def close(self):
self.__net.end_process()
class RunnerSignal(QObject):
finished = pyqtSignal()
class Mnist(Process):
def __init__(self, queue: Queue):
Process.__init__(self)
self.__queue = queue
def run(self):
mnist = tf.keras.datasets.mnist # 28x28 Bilder hangeschriebener Ziffern von 0-9
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = tf.keras.utils.normalize(x_train, axis=1)
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128, activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(128, activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(10, activation=tf.nn.softmax))
model.compile(optimizer="adam",
loss="sparse_categorical_crossentropy",
metrics=['accuracy'])
model.fit(x_train, y_train, epochs=8)
self.__queue.put(NetSignal.finished)
self.__queue.close()
def end_process(self):
self.terminate()
self.__queue.put(NetSignal.finished)
class NetSignal:
finished = "finished"
if __name__ == "__main__":
main_thread = QApplication(argv)
main_window = Window()
main_window.show()
exit(main_thread.exec())
