我在试图建立一个有线电视新闻网进行分类水果。 我一直遇到高损失的价值观和我在试图降低它作为我可以但是我不确定如何改进我的模型进一步。
这里是我的代码:
model96 = tf.keras.Sequential()
#Architecture
model96.add(tf.keras.layers.Conv2D(filters = 32,
kernel_size = (3, 3),
activation = "relu",
input_shape = (96, 96, 3)))
model96.add(tf.keras.layers.Conv2D(filters = 32,
kernel_size = (3, 3),
activation = "relu"))
model96.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))
model96.add(tf.keras.layers.Dropout(rate=0.25))
model96.add(tf.keras.layers.Flatten())
model96.add(tf.keras.layers.Dense(units=128, activation='relu'))
model96.add(tf.keras.layers.Dropout(rate=0.5))
#output layer
model96.add(tf.keras.layers.Dense(units=4, activation='softmax'))
#Loss function
model96.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
#Train model
hist96 = model96.fit(x=x_train96_norm, y=y_train, epochs=100)
#Test and Evaluate
print("Performance with test data:")
loss96, accuracy96 = model96.evaluate(x=x_test96_norm, y=y_test)
print('loss =', loss96)
print('accuracy =', accuracy96)
在培训过程中,最终损失的价值是0.0153和最终的精度值是0.9958,但是,在测试模型分: loss = 1.0462701320648193
和 accuracy = 0.8666666746139526