如何将整数与np.array一起保存

Question

I am doing a ConvNet and need to save one image alongside an integer (1 or 0), that indicates if an action was taken or not. How can I do it? I tried this:

key_pressed = np.array(key_check(self.key)) # key_check() returns 1 or 0
window = np.array(window) # window is opened with cv2

print(window.shape, key_pressed.shape) # (474, 31) (1,)
print(np.concatenate((window, key_pressed)))
      ^^
ValueError: all the input arrays must have same number of dimensions

Is doing what I want even possible with numpy, seeing that the arrays are of different sizes? Any ideas on how to save this data?

Answer 1

NumPy won't let you simply concatenate things of arbitrary sizes: that breaks the functional definition of an array or matrix. You're trying to make a simple juxtaposition, not a functional concatenation. All you're trying to do is display them side by side, right? This isn't a matrix operation. For instance, do you really envision doing a matrix multiplication on the resulting glued array???

You need a simple container to hold both items. I expect that a trivial class will do the job for you. Write a combined display method for the resulting pair, and use that for your rendering.

Answer 2

Have a look at scipy.io.savemat and loadmat :

savemat(your_path, dict(key=key_pressed, window=window))

Then using loadmat you'll get back a dict with the key and the frame that you want. There is even a do_compression parameter.

Edit: the incompatibility in shapes have nothing to do with it, since you're not creating a new array but saving key and window separately in the same file.

In [53]: from scipy import io

In [54]: a = np.zeros([3, 4])

In [55]: b = np.ones([5, 17])

In [56]: io.savemat("foo.mat", dict(a=a, b=b))

it works like a charm.