keras梯度计算失败“来自维度的无效索引:3、0，C"； [英] keras gradients computation fails "Invalid index from the dimension: 3, 0, C"

问题描述

当我尝试在具有conv1d层的网络中计算X wrt Y的梯度(X或Y是什么都无关紧要)时，我收到消息维度无效索引:3、0， C"，该过程终止.

When I try to compute the gradient of X w.r.t Y (doesn't really matter what X nor Y is) in a network with a conv1d layer I'm getting the message "Invalid index from the dimension: 3, 0, C" and the process dies.

最小工作示例:

import numpy as np

from tensorflow.python.keras import models
from tensorflow.python.keras import layers
from tensorflow.python.keras import backend as K

inp = layers.Input(shape=(10, 20,))
conv = layers.Conv1D(filters=10, kernel_size=2)(inp)
pool = layers.GlobalMaxPool1D()(conv)
output = layers.Dense(1, activation="sigmoid")(pool)

m = models.Model(inp, output)

m.summary()

m.compile(optimizer="adam", loss="binary_crossentropy")

似乎可以工作:

m.fit(x=np.random.randn(100, 10, 20), y=np.random.randn(100))

此中断:

loss = K.mean(m.output)
grads = K.gradients(loss, m.input)[0]
f = K.function([m.input], [grads])
print(f([np.random.randn(10, 20)]))

我的python，keras，tf版本:

My python, keras, tf versions:

import tensorflow as tf
import sys
from tensorflow.python import keras

print(tf.__version__)
print(keras.__version__)
print(sys.version)

1.12.0
2.1.6-tf
3.6.7 |Anaconda, Inc.| (default, Oct 23 2018, 14:01:38) 
[GCC 4.2.1 Compatible Clang 4.0.1 (tags/RELEASE_401/final)]

与计算梯度到什么无关紧要..错误消息是

It doesn't really matter of what I calculate the gradient wrt to what.. The error message is

2019-04-19 17:00:58.249788: F ./tensorflow/core/util/tensor_format.h:420] Check failed: index >= 0 && index < dimension_attributes.size() Invalid index from the dimension: 3, 0, C

我看到它与基于错误消息的转换1d层有关，但我不太了解我在这里缺少的内容.谢谢你的提示.

I see that it is related to the conv 1d layer based on the error message, but I don't quite understand what I'm missing here. Thanks for any hints.

推荐答案

简短答案:形状不兼容，将调用更改为:f([np.random.randn(1, 10, 20)]).

Short answer: shape incompatibility, change the call to: f([np.random.randn(1, 10, 20)]).

长答案::由于您将输入形状设置为(10, 20,)，这意味着每个输入样本的形状为(10,20).但是，您还必须注意，Keras模型需要一批样品作为输入.因此，在这种情况下，它将期望一个具有3维的数组，其中第一个维表示批处理维.由于您要使用一个样本来填充模型，因此输入数组的形状必须为(1, 10, 20).因此，您必须在randn函数中相应地更改形状:

Long answer: Since you have set the input shape as (10, 20,), it means that each input sample has a shape of (10,20). However, you must also note that Keras models expect a batch of samples as their input. Therefore, in this case it would expect an array with 3 dimensions where the first dimension indicates the batch dimension. Since you want to feed the model with one sample, the input array must have a shape of (1, 10, 20). So you must change the shape in randn function accordingly:

f([np.random.randn(1, 10, 20)])
                   ^
                   |
                   |
              batch dimension

这篇关于keras梯度计算失败“来自维度的无效索引:3、0，C"；的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！