Connecting jupyter widgets with QA pipeline for question selection problem

Question

I am using one of the pretrained language models from Huggingface for Question Answering but I am struggling to connect the jupyter widgets with the nlp pipeline.

I have this snippet which shows the user a set of questions that they can use for their QA.

w = widgets.Dropdown(
  options=['What are the risks of defamation?', 'Are there risks?', \
'What are the risks of the trial?', \
'Is the trial risky?'],
  value='What are the risks of defamation?',
  description='Question:',
)


def on_change(change):
    if change['type'] == 'change' and change['name'] == 'value':
        trial_q = change['new']
        #print(trial_q) works fine
        

w.observe(on_change)
display(w)

The selection works perfectly fine. However, I have trouble connecting it with the pipeline.

def QA(q, a):
    print(nlp({
    'question': q,
    'context': a
}))
    
def transcriptName():

  with open('data/uploaded_file.txt', 'r') as myfile: 
  data=myfile.read().replace('\n', '')

trial_q = QA('What are you randomised for?', data)

When calling the function, I get ValueError: Input is not valid. Should be a string, a list/tuple of strings or a list/tuple of integers.

So I assume it would be a problem with the context. However, if I change the trial_q inside the function with the string representation or 'What is the risk?', it works perfectly fine. What could be the issue?

Answer 1

It was a scope issue. The scope is within the widget.

w = widgets.Dropdown(
  options=['What are the risks of defamation?', 'Are there risks?', \
'What are the risks of the trial?', \
'Is the trial risky?'],
  value='What are the risks of defamation?',
  description='Question:',
)


def on_change(change):
    if change['type'] == 'change' and change['name'] == 'value':
        trial_q = change['new']
        
        ...pipeline here...
        

w.observe(on_change)
display(w)