Something Wrong

[charlesfufu]

[ShomyLiu]

Hi, file is only in Python 2.7X, and your python version is 3.6. You can try replace file into open

[charlesfufu]

Hi, file is only in Python 2.7X, and your python version is 3.6. You can try replace file into open

How about this one?

[ShomyLiu]

The training phase should be in this way in the terminal:

python main_sem.py  train

If you run code in IDE (e.g., Pycharm) , You can modify the __main__ function with:

if __name__ == "__main__":
     train()

[charlesfufu]

The training phase should be in this way in the terminal:

python main_sem.py  train

If you run code in IDE (e.g., Pycharm) , You can modify the __main__ function with:

if __name__ == "__main__":
     train()

According to what you say, get the same result.

[ShomyLiu]

The normal output of main_sem should be:

Maybe the environment difference?

[charlesfufu]

def train(**kwargs):

Where is the value of the parameter?

[charlesfufu]

The normal output of main_sem should be:

Maybe the environment difference?

Maybe you should upgrade your program with python3

[charlesfufu]

The normal output of main_sem should be:

Maybe the environment difference?

Maybe you should upgrade your program with python3

I don't know if you write program like these *** if name == "main": train()***,how to pass in the value at def train(**kwargs) function ?

[ShomyLiu]

OK, I will update my code with Python3.X.
You can change config.py Line 66:

.iteritems()   ->   .items()
```
Maybe there will be other codes different from python2.x and 3.x

[charlesfufu]

OK, I will update my code with Python3.X.
You can change config.py Line 66:

.iteritems()   ->   .items()

Maybe there will be other codes different from python2.x and 3.x

Yes, I had found at least four difference between python2 and python3. Looking forward to your code with python3. Thanks! !

[ShomyLiu]

Yes, I'm trying to update with python3. And I think it would be better with Pytorch 0.4 or later in the revision.

[charlesfufu]

The normal output of main_sem should be:

Maybe the environment difference?

Maybe you should upgrade your program with python3

I don't know if you write program like these *** if name == "main": train()***,how to pass in the value at def train(**kwargs) function ?

I can not run the part program of picture. Do you know what's reason causes these? my pytorch version is 0.4

[ShomyLiu]

Yeah. My code is in Pytorch0.3. I'm updating code into Python3.x and Pytorch 0.4

[ShomyLiu]

@charlesfufu The code has updated. the master branch is under Python 3.5x and Pytorch 0.4+.

[charlesfufu]

@charlesfufu The code has updated. the master branch is under Python 3.5x and Pytorch 0.4+.

Wonderful !!

[charlesfufu]

@charlesfufu The code has updated. the master branch is under Python 3.5x and Pytorch 0.4+.

Wonderful !!

Do you know why?

[Wangpeiyi9979]

try changing the "use_gpu" in config.py to False.

[charlesfufu]

try changing the "use_gpu" in config.py to False.

still wrong

[Wangpeiyi9979]

Is the same wrong message?

[charlesfufu]

Is the same wrong message?

yeah！！Maybe is not GPU problem.

[Wangpeiyi9979]

try changing the codes here to this

[charlesfufu]

try changing the codes here to this

No,that's wrong.

[Wangpeiyi9979]

this?

[Wangpeiyi9979]

also change the code here, and add long(). it may work

[charlesfufu]

No, not right
Can I ask you about it, have you run it since the program was modified?

[Wangpeiyi9979]

[charlesfufu]

@Wangpeiyi9979 what about win10 ?

[Wangpeiyi9979]

@Wangpeiyi9979 what about win10 ?
it should work. changing the code here, add long()..

and here

[charlesfufu]

@ShomyLiu

[ShomyLiu]

@charlesfufu I have reviewed code carefully and cannot reproduce your problems either in GPU or CPU.
Please check your environments.
My environments are listed here:

- OS: Ubuntu 16.04 
- Python: 3.5
- Pytorch: 0.4.1
- GPU: NVIDIA GTX 1080Ti with 387.34 drivers.

[charlesfufu]

Have you tested your program on windows?I think maybe is the operating system problem

[charlesfufu]

I had found the bug

@charlesfufu I have reviewed code carefully and cannot reproduce your problems either in GPU or CPU.
Please check your environments.
My environments are listed here:

- OS: Ubuntu 16.04 
- Python: 3.5
- Pytorch: 0.4.1
- GPU: NVIDIA GTX 1080Ti with 387.34 drivers.

I had found the bug ! !

[ShomyLiu]

OK, It seems strange. It would be nice If the issue has been solved. If necessary, you can create a pull request.
Thanks.

[ghost]

I had found the bug

@charlesfufu I have reviewed code carefully and cannot reproduce your problems either in GPU or CPU.
Please check your environments.
My environments are listed here:

- OS: Ubuntu 16.04 
- Python: 3.5
- Pytorch: 0.4.1
- GPU: NVIDIA GTX 1080Ti with 387.34 drivers.

I had found the bug ! !

你好，我遇到和你一样的报错信息【上图】，但没找到应该在哪里修改这个bug。看起来是pytorch在对数据类型进行转换的时候出现的问题，请问你是如何解决的呢？最后整个模型可以跑起来了吗？

[XAKK]

I had found the bug

@charlesfufu I have reviewed code carefully and cannot reproduce your problems either in GPU or CPU.
Please check your environments.
My environments are listed here:

- OS: Ubuntu 16.04 
- Python: 3.5
- Pytorch: 0.4.1
- GPU: NVIDIA GTX 1080Ti with 387.34 drivers.

I had found the bug ! !

Can you show the details about fixing this bug?

[ShomyLiu]

@XAKK
Hi, could you please show what's the problem here?
I have checked the codes and run successfully right now under the following environment:

• Ubuntu 18.04
• Python 3.X
• Pytorch 0.4+ (1.0 is ok)

[XAKK]

@ShomyLiu
Hi, I have the same problem as @charlesfufu
it seems that this problem only appear on the Windows.

Traceback (most recent call last):
File "main_sem.py", line 123, in
fire.Fire()
File "C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python36_64\lib\site-packages\fire\core.py", line 127, in Fire
component_trace = _Fire(component, args, context, name)
File "C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python36_64\lib\site-packages\fire\core.py", line 366, in _Fire
component, remaining_args)
File "C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python36_64\lib\site-packages\fire\core.py", line 542, in _CallCallable
result = fn(*varargs, **kwargs)
File "main_sem.py", line 64, in train
out = model(data[:-1])
File "C:\Users\79052\AppData\Roaming\Python\Python36\site-packages\torch\nn\modules\module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "D:\SFOH\pytorch-pcnn-master\models\PCNN.py", line 72, in forward
lexical_level_emb = self.word_embs(lexical_feature) # (batch_size, 6, word_dim
File "C:\Users\79052\AppData\Roaming\Python\Python36\site-packages\torch\nn\modules\module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "C:\Users\79052\AppData\Roaming\Python\Python36\site-packages\torch\nn\modules\sparse.py", line 118, in forward
self.norm_type, self.scale_grad_by_freq, self.sparse)
File "C:\Users\79052\AppData\Roaming\Python\Python36\site-packages\torch\nn\functional.py", line 1454, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: Expected tensor for argument #1 'indices' to have scalar type Long; but got torch.cuda.IntTensor instead (while checking arguments for embedding)

[ShomyLiu]

@XAKK
I will reproduce the problem in Windows platform as soon as possible and find out the solution.

[XAKK]

I think I found a problem solving strategy for Windows platform.
I modified the codes for three places, and run successfully on

• Windows 10

• Python 3.6

• Pytorch 1.0

Here are the changes:
main_sem.py
line 65:
loss = criterion(out, data[-1])
↓
out1 = out.type(torch.cuda.FloatTensor) data1 = data[-1].type(torch.cuda.LongTensor) loss = criterion(out1, data1)

line 92 - 95:
if opt.use_gpu: data = list(map(lambda x: torch.LongTensor(x).cuda(), data)) else: data = list(map(lambda x: torch.LongTensor(x), data))
↓

if opt.use_gpu: # data = list(torch.cuda.LongTensor(long(i)) for i in data) data = list(map(lambda x: torch.LongTensor(x.long()).cuda(), data)) else: data = list(map(lambda x: torch.LongTensor(x.long()), data))

PCNN.py line 68:
lexical_feature, word_feautre, left_pf, right_pf = x
↓
lexical_feature, word_feautre, left_pf, right_pf = x lexical_feature = lexical_feature.type(torch.cuda.LongTensor) word_feautre = word_feautre.type(torch.cuda.LongTensor) left_pf =left_pf.type(torch.cuda.LongTensor) right_pf = right_pf.type(torch.cuda.LongTensor)

This is part of my result with defualt setting

<<< (9+1)-WAY EVALUATION TAKING DIRECTIONALITY INTO ACCOUNT -- OFFICIAL >>>:
Micro-averaged result (excluding Other):
P = 1907/2409 = 79.16% R = 1907/2263 = 84.27% F1 = 81.64%

MACRO-averaged result (excluding Other):
P = 78.58% R = 83.26% F1 = 80.77%
Thank you! @ShomyLiu

[ShomyLiu]

@XAKK Thanks for the note.
I will refine the codes for windows recently.
It seems that the code above can be simplified with:

 out.type(torch.cuda.FloatTensor) -> out.float()
...

lexical_feature.type(torch.cuda.LongTensor) -> lexical_feature.long()

[Xin-Deng]

我并没有修改任何东西 目录下的确有该文件 为什么会找不到呢？

[ShomyLiu]

@learning-makes-me-happy-a 请确认下执行命令所在的目录。 需要在主目录，而不是说dataset下

[Xin-Deng]

@learning-makes-me-happy-a 请确认下执行命令所在的目录。 需要在主目录，而不是说dataset下

thanks I have solved this problem