分类形状中的keras ValueError

Question

这是我的代码：

# data consists of 1 dimensional data (3277 elements). Number of data is 439  
train_data = .... # numpy.ndarray
# I would like to classify data into 5 classes.
train_labels = .... # numpy.ndarray

print(train_data.shape) # -> Shape of train_data: (439, 3277)
print('Shape of train_labels:', train_labels.shape) # -> Shape of train_labels: (439,)
# prepare 5 one hot encoding array
categorical_labels = to_categorical(train_labels, 5)
print('Shape of categorical_labels:', categorical_labels.shape) # -> Shape of categorical_labels: (439, 5)

# I make a model to have 3277-elements data and classify data into 5 labels.
model = keras.Sequential([
    keras.layers.Dense(30, activation='relu', input_shape=(3277,)),
    keras.layers.Dense(30, activation='relu'),
    keras.layers.Dense(5, activation='softmax')
])
model.summary()
model.compile(optimizer='adam',
          loss='sparse_categorical_crossentropy',
          metrics=['accuracy'])
model.fit(data, categorical_labels, epochs=5, verbose=1) # A
#model.fit(data, train_labels, epochs=5, verbose=1) # B

当我尝试使用标记为 'A' 的行时，出现此错误

ValueError: Error when checking target: expected dense_3 to have shape (1,) but got array with shape (5,)

使用'B'，运行正常（无明显错误，机器returns高分）

显然，错误与形状的差异有关...当我想使用 keras.utils.to_categorical 时，我该如何修改我的代码？

另一个问题是为什么这个案例有效（https://github.com/keras-team/keras/blob/master/examples/mnist_mlp.py）而我的案例不...

结构看起来很像……对我来说。

Answer 1

因为sparse_categorical_crossentropy不希望使用one-hot编码格式的标签，所以你应该使用loss='categorical_crossentropy'。

简而言之，关于你的情况：

train_labels => loss='sparse_categorical_crossentropy'
categorical_labels => loss='categorical_crossentropy'

分类形状中的keras ValueError

keras ValueError in classification shape

python

classification

machine-learning

keras

tensorflow