lyeoni / nlp-tutorial Goto Github PK

Hi
After saving the model in news-category-classification, how do you actually use it to predict text classification?
Can you put up an example, please?

Thanks for your code!

I found that when I do training, the GPU are not totally utilized. So it there is way to add batch to train more pairs at one iter?

Hi, I am curious about the inference part in the model. Does any example to show how it works? Many thank.

Hi Lyeoni,

First of all, thank you a lot for your work in making these tutorials, which are interesting !

I am trying to run the question-answer-matching tutorial and reproduce your evaluation. Unfortunately, I can't download the Posts.xml file from git lfs as it looks like your subscription doesn't accept download anymore.
By any chance, do you have that file hosted somewhere else ? That would allow me to run the evaluation with your trained model.

Thanks a lot and I wish you a nice day ! :-)

Kindly add a tutorial for NLP with transformer setup

@lyeoni

In the case I want to train an Arabic to Urdu Machine Translation:

• is that attainable using this project?
• what options should be set in training?
• do you suggest another github project?

Hi,
In cleaning function in the script : nlp-tutorial/news-category-classifcation/preprocessing.py,
line 21 is written as text = re.sub(r'[!]{2,}', '?', text) # multiple ?s -> ?. There should be ? in first argument and It should be text = re.sub(r'[?]{2,}', '?', text) # multiple ?s -> ?.
Am I correct?

Hi, @lyeoni !
You have written great tutorials. I really appreciate you)
We can improve a little bit with one pretty line. Look, please)
Here, we fill first key-value items of stoi, itos by special tokens.
I suggest insert this line before cycle.
special_tokens = filter(lambda x: x is not None, [self.unk_token, self.bos_token, self.eos_token, self.pad_token])
If we don't set value for self.unk_token and set for self.bos_token, then index in dictionary become wrong. So, we need filter None values before.
Input
vocab = Vocab(body, bos_token='<bos>'); vocab.build(); vocab.stoi;
Wrong Output
'<bos>': 1 ' ': 1, 'hi': 2, 'bear': 3, ...

python train.py --epochs 12 --batch_size 2 --learning_rate .001 --hidden_size 64 --n_layers 1 --dropout_p .1

number of trained word vectors of data/glove.6B.100d.txt: 400000
Traceback (most recent call last):
File "train.py", line 200, in
train_loader = DataLoader(dataset=qa_train, batch_size=config.batch_size, shuffle=True, num_workers=4)
File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataloader.py", line 176, in init
sampler = RandomSampler(dataset)
File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/sampler.py", line 66, in init
"value, but got num_samples={}".format(self.num_samples))
ValueError: num_samples should be a positive integer value, but got num_samples=0

Please Let Know What's the exact issue

Traceback (most recent call last):

File "", line 1, in
runfile('D:/nlp-tutorial/neural-machine-translation/nmt/train.py', wdir='D:/nlp-tutorial/neural-machine-translation/nmt')

File "C:\ProgramData\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 827, in runfile
execfile(filename, namespace)

File "C:\ProgramData\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 110, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "D:/nlp-tutorial/neural-machine-translation/nmt/train.py", line 254, in
trainiters(pairs, encoder, decoder, n_iters)

File "D:/nlp-tutorial/neural-machine-translation/nmt/train.py", line 184, in trainiters
train_pairs += [random.choice(train_pairs) for i in range(n_iters%len(train_pairs))]

ZeroDivisionError: integer division or modulo by zero

Movie Rating Classification no datasets?

The corresponding package is installed and Data set downloaded，Run vocab.py . The following error occurred： “local variable 'MosesTokenizer' referenced before assignment”

Thanks for your great job! I learned a lot. However, I have a question.
I train the model for 7 epochs reaching a train acc of 95.2 and test(validate) acc of 85.2.
Is that normal? Could the final test(validate) acc be higher after more epochs? Thanks!

I had install nltk; but a error occur when I run the code;
ModuleNotFoundError: No module named 'nltk.tokenize.moses'