CODEWITHSUNDEEP
pythonmultidimensional-arraytensorflow
Emanuela Liaci
Emanuela Liaci

Reputation: 185

series of 2D arrays with different dimension lengths in feed_dict

I have a series of 2d arrays which always differ for one dimension size, e.g. (20,87), (20,100), (20,76), etc... Those arrays are composed by Mel-frequency cepstrum coefficients (mfccs) for time steps (times), so (mfccs, times).

In oder to train a CNN in Tensorflow, I need to feed a dictionary with a batch of some of those 2d arrays.

I would like to have a 3d array because my input tensor would be: x=tf.placeholder('float', shape=(n,mfccs, times)), where n is the batch size. So the batch would be a 3d array with this shape: (n,mfccs,times), where only the size of times dimension changes.

I thought also to use a list of 2d arrays instead of a 3d array. But is it possible to feed a list in the feed_dict (e.g. feed_dict={ x: list?})? if yes, how do you do that?

Many thanks for the help in advance.

Upvotes: 1

Views: 557

Answers (1)

Dawid Laszuk
Dawid Laszuk

Reputation: 1978

For different sizes, but acting as an array, one can try list:

a = [[0]*87 for range(20)]
b = [[0]*100 for range(20)]
c = [[0]*76 for range(20)]

big_list = []
big_list.append(a)
big_list.append(b)
big_list.append(c)

After all this, big_list has length of 3, where each elements contains respective list/array.

Upvotes: 0

Related Questions