keras - 当我添加 step_per_epoch 且 epoch=1 时，训练 ETA 为 5 小时

Question

traindatagenerator = imagegenerator.flowfromdataframe(traindf, directory = IMAGEDIR,xcol = "imagename", ycol = "target",classmode = "raw", batchsize=25, targetsize=(240,240),shuffle = True) validdatagenerator = imagegenerator.flowfromdataframe( valdf, directory = IMAGEDIR,xcol = "imagename", ycol = "target",classmode = "raw", batchsize=25, targetsize=(240,240),shuffle = True) classweights = classweight.computeclassweight('balanced',np.唯一的（traindatagenerator.labels），traindatagenerator.labels）

创建模型 model=get_model()

编译模型 model.compile(loss='binary_crossentropy',optimizer ='sgd',metrics=[tf.keras.metrics.AUC()])

创建回调

checkpoint = keras.callbacks.ModelCheckpoint('model'+str(foldvar)+'.h5',monitor='valaccuracy', verbose=1,savebest_only=True, mode='max')

callbacks_list = [检查点]

history = model.fitgenerator(traindatagenerator,validationdata=validdatagenerator,stepsperepoch=int(math.ceil(1. * Xtrain.shape[0] // 25)),validationsteps=int(math.ceil(1. * XVal.shape[ 0] // 25)),callbacks=callbackslist,classweight=classweights,epochs =1)

标签： kerasresnet