numpy.random.shuffle
在做将caffe模型和预训练的参数转化为tensorflow的模型和预训练的参数,以便微调,遇到如下函数:
def gen_data(source):
while True:
indices = range(len(source.images)) # indices = the number of images in the source data set
random.shuffle(indices)
for i in indices:
image = np.reshape(source.images[i], (28, 28, 1))
label = source.labels[i]
yield image, label
之前卑鄙陋寡闻,不知道这个用法,按照字面上的意思是打乱,那么这里就应该是让训练数据集中的数据打乱顺序,然后一个挨着一个地(for i in indices)生成训练数据对。下面就从docs.scipy.org中查到的random.shuffle的用法:
numpy.random.shuffle(x)
Modify a sequence in-place by shuffling its contents.
Parameters: |
x : array_like
|
---|---|
Returns: |
None |
举例
python>>>
>>> arr = np.arange(10)
>>> np.random.shuffle(arr)
>>> arr
[1 7 5 2 9 4 3 6 0 8]
This function only shuffles the array along the first index of a multi-dimensional array(多维矩阵中,只对第一维(行)做打乱顺序操作):
python>>>
>>> arr = np.arange(9).reshape((3, 3))
>>> np.random.shuffle(arr)
>>> arr
array([[3, 4, 5],
[6, 7, 8],
[0, 1, 2]])This function only shuffles the array along the first index of a multi-dimensional array:
参考:
[1] https://docs.scipy.org/doc/numpy/reference/generated/numpy.random.shuffle.html#numpy-random-shuffle
[2] https://github.com/ethereon/caffe-tensorflow/blob/master/examples/mnist/finetune_mnist.py