Keras因其形状而不接受目标标签?

Question

我正在尝试让 Keras 处理一个分类问题，该问题有五个分类目标标签(1、2、3、4、5(。由于某种原因，我无法使用StratifiedKFold使其工作。X 和 y 分别是形状为 (500， 20( 和 (500， ( 的 NumPy 数组。

错误消息是"ValueError：检查目标时出错：预期dense_35具有形状 (1，(，但得到具有形状 (5，(的数组"，这让我认为错误肯定在于目标变量的格式。同样值得注意的是，每次尝试运行代码时，"dense_35"中的数字似乎都在变化。

random_state = 123
n_splits = 10
cv = StratifiedKFold(n_splits=n_splits, 
random_state=random_state, shuffle=False)
def baseline_model():
nn_model = Sequential()
nn_model.add(Dense(units=50, input_dim=X.shape[1], init='normal',
activation= 'relu' ))
nn_model.add(Dense(30, init='normal', activation='relu'))
nn_model.add(Dense(10, init='normal', activation='relu'))
nn_model.add(Dense(1, init='normal', activation='softmax'))
nn_model.compile(optimizer='adam', loss='categorical_crossentropy',
metrics = ['accuracy'])
return nn_model
for train, test in cv.split(X, y):   
X_train, X_test = X[train], X[test]
y_train, y_test = y[train], y[test]
np_utils.to_categorical(y_train)
np_utils.to_categorical(y_test)
estimator = KerasClassifier(build_fn=baseline_model,
epochs=200, batch_size=5,
verbose=0)
estimator.fit(X_train, y_train)
y_pred = estimator.predict(X_test)
The numpy array (y), that I am trying to split:
[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5]

Answer 1

random_state = 123
n_splits = 10
cv = StratifiedKFold(n_splits=n_splits, random_state=random_state, shuffle=False)
def baseline_model():
nn_model = Sequential(name='model_name')
nn_model.add(Dense(units=50, input_dim=X.shape[1], init='normal',
activation= 'relu', name='dense1'))
nn_model.add(Dense(30, init='normal', activation='relu', name='dense2'))
nn_model.add(Dense(10, init='normal', activation='relu', name='dense3'))
# code changed here
nn_model.add(Dense(5, init='normal', activation='softmax', name='dense4'))
nn_model.compile(optimizer='adam', loss='categorical_crossentropy',
metrics = ['accuracy'])
return nn_model
for train, test in cv.split(X, y):   
X_train, X_test = X[train], X[test]
y_train, y_test = y[train], y[test]
# the error is due to this step
# you have specified only one output in the last dense layer (dense4)
# but you are giving input of length 5
np_utils.to_categorical(y_train)
np_utils.to_categorical(y_test)
estimator = KerasClassifier(build_fn=baseline_model,
epochs=200, batch_size=5,
verbose=0)
estimator.fit(X_train, y_train)
y_pred = estimator.predict(X_test)

• 通过在图层中指定名称参数，可以命名图层。通过这样做，您将在每次出错的情况下获得图层的明确名称。
• model.summary((是另一个有用的函数，您可以使用它检查每个图层的输出形状。

Answer 2

由于这是一个分类问题，它有五个分类目标标签，最后一个密集层(输出层(必须有 5 个单位：

def baseline_model():
nn_model = Sequential()
nn_model.add(Dense(units=50, input_dim=X.shape[1], init='normal',
activation= 'relu' ))
nn_model.add(Dense(30, init='normal', activation='relu'))
nn_model.add(Dense(10, init='normal', activation='relu'))
#Output layer
nn_model.add(Dense(5, init='normal', activation='softmax'))
nn_model.compile(optimizer='adam', loss='categorical_crossentropy',
metrics = ['accuracy'])
return nn_model