Why do I get "TypeError: expected np.ndarray (got numpy.ndarray)" when I use torch.from_numpy() function?

1

Why do I get “TypeError: expected np.ndarray (got numpy.ndarray)” when I use torch.from_numpy() function? Isn’t np.ndarray equivalent to numpy.ndarray? Also, there doesn’t seem to be any np.ndarray type, but only numpy.ndarray type.

Traceback (most recent call last): File "test_opencv.py", line 31, in <module> bounding_boxes, landmarks = detect_faces(img, pnet, rnet, onet) File "/home/xxx/git/project/src/detector.py", line 67, in detect_faces boxes = run_first_stage(image, pnet, scale=s, threshold=thresholds[0]) File "/home/xxx/git/project/src/first_stage.py", line 73, in run_first_stage img = torch.from_numpy(img).unsqueeze(0).permute(0,3,1,2).cuda().float() TypeError: expected np.ndarray (got numpy.ndarray)

2

Can you share some debugging info like printing type(img) and img.dtype I wrote the same line of code and it worked for me.
There has to be some bug in the data loading process of the image.

3

Hello It seems you are getting the error because the argument to from_numpy function is a single value rather than array.
In numpy, there is difference between np.array(1) and np.array([1]) and both are completely different data types.
Try torch.from_numpy(np.asarray(x)).

5 Likes
4

Hello, I have encountered the same problem. Do you have any solutions?Thank you

5

Try
torch.as_tensor(np.array(pil_img).astype('float'))

It worked for me.

2 Likes
6

Have you fix this problem? I encounter this problem too…

7

Could you post a minimal code snippet to reproduce this error?

8

I found it is because of the version of numpy…




信工 刘蓬博

邮箱：pengbo18555@163.com

签名由 网易邮箱大师 定制

9

It’s also happening to me on numpy 1.17.4, my own solution is to downgrade to 1.16.4

10

thanks! it worked for me.
Numpy version - 1.19.5

11

It worked for me. Thank you @raghavendragaleppa

12

I was working with tabular MNIST data !!
one possible solution that worked for me was,

x_train.reset_index(drop=True,inplace=True)
x_test.reset_index(drop=True,inplace=True)
y_train.reset_index(drop=True,inplace=True)
y_test.reset_index(drop=True,inplace=True)

hope it helps for those using TABULAR dataset

13

iam trying to run a simple dqn code for cartpole environment in my computer. This is my code updated as you said;

    env = gym.make('CartPole-v1')
    q = Qnet()
    q_target = Qnet()
    q_target.load_state_dict(q.state_dict())
    memory = ReplayBuffer()

    print_interval = 20
    score = 0.0  
    optimizer = optim.Adam(q.parameters(), lr=learning_rate)

    for n_epi in range(10000):
        epsilon = max(0.01, 0.08 - 0.01*(n_epi/200)) #Linear annealing from 8% to 1%
        s = env.reset()
        done = False
        while not done:
            **a = q.sample_action(torch.from_numpy(np.asarray(s)).float(), epsilon)**      
            s_prime, r, done, info = env.step(a)
            done_mask = 0.0 if done else 1.0
            memory.put((s,a,r/100.0,s_prime, done_mask))
            s = s_prime

but iam getting the following error again where everything is fine
Traceback (most recent call last):
File “d:\college\final yr\project\rl-series\dqn\simpledqn.py”, line 114, in
main()
File “d:\college\final yr\project\rl-series\dqn\simpledqn.py”, line 93, in main
a = q.sample_action(torch.from_numpy(np.asarray(s)), epsilon)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part.

14

Thanks! This solved my issue.

15

I think this problem may arise from torch.from_numpy(array)'s mismatching with numpy version or dtype,and use other way of generating a tensor from numpy is a solution.