我正在研究MNIST数据集,并决定尝试进行数据生成。现在直接:
我正在使用验证拆分生成数据,如下所示:
datagen = keras.preprocessing.image.ImageDataGenerator(... , validation_split=0.21, ...)
datagen.fit(train_x)
但是我不知道如何将这种验证拆分称为模型的拟合度
hist = model.fit_generator(datagen.flow(train_x, train_y, batch_size =32),
steps_per_epoch=len(train_x)//32,
epochs = 70, verbose=0, callbacks= [PlotLossesKeras()],
validation_data= **???**, <-----
validation_steps=None,
validation_freq=1,
class_weight=None,
max_queue_size=10,
workers=1,
use_multiprocessing=False,
shuffle=True,
initial_epoch=0)
我在文档或其他地方都找不到答案。你能帮我吗?
答案 0 :(得分:2)
您应说明您正在使用哪个子集:subset = 'validation'
,如Keras documentation所说:
子集:如果在ImageDataGenerator中设置了validate_split,则数据的子集(“训练”或“验证”)。
例如,您可以执行以下操作:
datagen = keras.preprocessing.image.ImageDataGenerator(..., validation_split=0.21)
train_generator = datagen.flow(..., subset='training')
valid_generator = datagen.flow(..., subset='validation')
hist = model.fit_generator(...,
generator = train_generator,
validation_data = valid_generator,
steps_per_epoch = len(train_generator),
validation_steps = len(valid_generator),
)
您还可以在ImageDataGenerator class
documentation