Array Size of MATLAB dlarray
While conducting deep learning in MATLAB, dlarray function1 is to convert the array type dataset to dlarray that available for custom training loop, and meanwhile, we could specify the data format by it. However, there is a small detail that should be noted, dlarray function will automatically permute the high-dimensional array when we specify the data format and the last dimension label is not "B" (denoting “batch” or rather “batch size”), to make the converted dlarray-type array the last dimension always "B". For example, the following two script show the difference of array size of specifying data format as "CB" and "BC" for two-dimensional array:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
clc,clear,close all
rng("default")
a = randn(3,10);
fprintf("size(a): %d %d\n",size(a))
a1 = dlarray(a); % Without specifying data format
fprintf("size(a1): %d %d, Dimension labels: %s\n",size(a1),dims(a1))
a2 = dlarray(a,"CB");
fprintf("size(a2): %d %d, Dimension labels: %s\n",size(a2),dims(a2))
a3 = dlarray(a,"BC");
fprintf("size(a3): %d %d, Dimension labels: %s\n",size(a3),dims(a3))
1
2
3
4
size(a): 3 10
size(a1): 3 10, Dimension labels:
size(a2): 3 10, Dimension labels: CB
size(a3): 10 3, Dimension labels: CB
dims function2 is to display the dimension labels of the input dlarray.
and that of specifying data format as "SSCB" and "SSBC" for four-dimensional array:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
clc,clear,close all
rng("default")
b = rand(3,3,10,7);
fprintf("size(b): %d %d %d %d\n",size(b))
b1 = dlarray(b); % Without specifying data format
fprintf("size(b1): %d %d %d %d, Dimension labels: %s\n",size(b1),dims(b1))
b2 = dlarray(b,"SSCB");
fprintf("size(b2): %d %d %d %d, Dimension labels: %s\n",size(b2),dims(b2))
b3 = dlarray(b,"SSBC");
fprintf("size(b3): %d %d %d %d, Dimension labels: %s\n",size(b3),dims(b3))
1
2
3
4
size(b): 3 3 10 7
size(b1): 3 3 10 7, Dimension labels:
size(b2): 3 3 10 7, Dimension labels: SSCB
size(b3): 3 3 7 10, Dimension labels: SSCB
References