Is there a way to reshape an array into a certain shape padding missing dimensions?
Let's say, we have an array x
x = np.arange(10)
Now we want to reshape it to (1, 1, 1, 10, 1). We can do:
x = x.reshape((1, 1, 1, 10, 1))
or
x = x[None, None, None, x.shape[0], None]
But if the number and position of padding dimensions vary, this becomes quite inconvenient. Is there a better way?
You can construct the new shape programatically:
pre_pad, post_pad = 5,2
new_shape = tuple([1]*pre_pad) + x.shape + tuple([1]*post_pad)
y = x.reshape(new_shape)
print(y.shape)
# (1, 1, 1, 1, 1, 10, 1, 1)
You could use np.pad, which gives you a lot of control:
x = np.arange(10)
new_shape = np.pad([x.shape[0]], # the thing to pad, in this case, [10]
(3,1), # padding on (left, right)
'constant',
constant_values=1) # could specify L/R instead, e.g. (2,3)
y = x.reshape(new_shape)
y.shape
# (1, 1, 1, 10, 1)