keras -为什么我的输出是nan？
keras -为什么我的输出是nan？
提问于 2016-07-13 16:30:43
关于任务:我将类距离作为输入，并希望获得类置信度(介于0.0和1.0之间的数字)。所以我有一些类似的东西：
[
[
0.0,
0.0,
0.0,
6.371921190238224,
0.0,
3.3287083713830516,
7.085957828217146,
7.747408965761948,
5.498717498872398,
5.498717498872398,
5.498717498872398,
5.498717498872398,
8.529725281060978
],
[
6.396501448825533,
0.0,
0.0,
5.217483270813266,
0.0,
5.319046151560534,
5.823161030197735,
3.8991256371824976,
6.269856323952211,
5.517874167220461,
6.396501448825533,
5.328678274963717,
3.8991256371824976
],
]结果是
[
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
...
]我有大约200个例子。下面是我的网络构建代码：
def train(self, distances, classes):
"""
Train network
:param distances: array of distances to classes
:type distances: list[list[float]]
:param classes: array of class indicators
:type classes: list[list[float]]
"""
example_count, class_count = self._dimensions(distances, classes)
self.model = Sequential()
self.model.add(Dense(128, input_dim=class_count))
self.model.add(Dense(class_count))
self.model.compile(optimizer=SGD(), loss='mse')
self.model.fit(array(distances), array(classes))但是在训练期间，我得到了下一个输出：
Epoch 1/10
425/425 [==============================] - 0s - loss: nan
Epoch 2/10
425/425 [==============================] - 0s - loss: nan
Epoch 3/10
425/425 [==============================] - 0s - loss: nan
Epoch 4/10
425/425 [==============================] - 0s - loss: nan
Epoch 5/10
425/425 [==============================] - 0s - loss: nan
Epoch 6/10
425/425 [==============================] - 0s - loss: nan
Epoch 7/10
425/425 [==============================] - 0s - loss: nan
Epoch 8/10
425/425 [==============================] - 0s - loss: nan
Epoch 9/10
425/425 [==============================] - 0s - loss: nan
Epoch 10/10
425/425 [==============================] - 0s - loss: nan 当我尝试使用model.predict(numpy.array([[ 0.0, 0.0, 0.0, 6.371921190238224, 0.0, 3.3287083713830516, 7.085957828217146, 7.747408965761948, 5.498717498872398, 5.498717498872398, 5.498717498872398, 5.498717498872398, 8.529725281060978]])) (来自训练集的示例)时，我得到了[[ nan nan nan nan nan nan nan nan nan nan nan nan nan]]
数据或构建代码中会出现什么错误？
回答
页面原文内容由Stack Overflow提供。腾讯云小微IT领域专用引擎提供翻译支持
原文链接：
https://stackoverflow.com/questions/38346448
复制相关文章
相似问题
腾讯云开发者
