删除具有自己的TensorFlow图的类不会将其从RAM存储器中删除

问题描述

我创建了一个类，该类在线程内部的循环中创建，训练和评估预测变量。我使用while循环运行线程，在该线程中创建该类以执行训练然后进行预测。运行循环后，最后删除预测变量类实例。 我的期望是，当我执行del predictor并跟进gc.collect()之后，实例中的tf图也应删除，并应从RAM内存中删除。但是每个循环的RAM使用量都在不断增加。

我正在使用tf 1.14.0和keras 2.2.4

import tensorflow as tf
from keras import backend as K
from keras.models import Model
from keras.layers import Input,Dense,LSTM,Reshape
from keras.callbacks import ModelCheckpoint,EarlyStopping,ReduceLROnPlateau,TensorBoard

import numpy as np

from multiprocessing import Event
from threading import Thread

import gc

class nn_model():
    """
    Creates the data driven model for predicting energy
    """

    def __init__(self,*args,**kwargs):
        """
        Initiate the class
        """
        self.graph = tf.Graph()
        self.session = tf.Session(graph=self.graph)

    def design_net(self,):
        with self.graph.as_default():  # pylint: disable=not-context-manager
            with self.session.as_default():  # pylint: disable=not-context-manager
                input_layer = Input(batch_shape=(None,1,3))
                layers = input_layer
                layers = Dense(64,activation='relu')(layers)
                layers = LSTM(32,activation='linear',return_sequences=True)(layers)
                output = LSTM(1,return_sequences=False)(layers)
                output = Reshape((1,1))(output)
                self.model = Model(inputs=input_layer,outputs=output)
                self.model.compile(loss=self.loss,optimizer='adam')
    def fit(self,**kwargs):
        """
        Fit the model to the data
        """
        # train the model
        with self.graph.as_default():  # pylint: disable=not-context-manager
            with self.session.as_default():  # pylint: disable=not-context-manager
                self.history = self.model.fit(kwargs['X_train'],kwargs['y_train'],validation_data=(kwargs['X_val'],kwargs['y_val']),epochs=50,callbacks=self.callbacks(),verbose=0)
    
    def callbacks(self,**kwargs):
        """
        Create callbacks
        """
        self.modelchkpt = ModelCheckpoint('best_model',monitor = 'val_loss',save_best_only = True,period=2)
        self.earlystopping = EarlyStopping(monitor = 'val_loss',patience=5,restore_best_weights=False)
        self.reduclronplateau = ReduceLROnPlateau(monitor = 'val_loss',patience=2,cooldown = 3)
        self.cb_list = [self.modelchkpt,self.earlystopping,self.reduclronplateau]
        return self.cb_list

    def predict(self,**kwargs):
        """
        Evaluate the model; pass args,kwargs if needed
        """
        with self.graph.as_default():  # pylint: disable=not-context-manager
            with self.session.as_default():  # pylint: disable=not-context-manager
                predictions = self.model.predict(kwargs['X_test'])
        return predictions

def learn_on_loop(end_looping,X_train,y_train,X_val,y_val):
    while not end_looping:

        predictor = nn_model()
        predictor.design_net()
        predictor.fit(**{'X_train':X_train,'y_train':y_train,'X_val':X_val,'y_val':y_val})
        np.save('y_pred.npy',predictor.predict(**{'X_val':X_val}))

        del predictor
        gc.collect()

        
if __name__ == "__main__":
    end_looping = Event()
    X_train,y_val = np.load('X_train.npy'),np.load('y_train.npy'),np.load('X_val.npy'),np.load('y_val.npy')
    predictor_th = Thread(target=learn_on_loop,daemon=False,kwargs={
        'end_looping':end_looping,'X_train':X_train,'y_val':y_val
    })
    predictor_th.start()
    try:
        predictor_th.join()
    except KeyboardInterrupt:
        end_looping.set()

this可能是内存仍在增加的原因吗？

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）