app.py

import streamlit as st
import datasets
from datasets import load_dataset
import pandas as pd
from streamlit.components.v1 import html
from streamlit import markdown
import re
import os
import time

import json


st.title('StackOverflow Question Demo')

library = st.radio('Select a library', ('numpy', 'tensorflow+ pytorch', 'scipy', 'scikit-learn', 'pandas'))

question_path = './{}.txt'.format(library)
# loading stackoverflow questions.
# using huggingface load_dataset function.
# not done yet
#@st.cache
#def load_data(path):
#    return load_dataset('text', data_files = path, cache_dir = './data')

intro = {'numpy': '''
#### Setup
temperature = 0.7, topP = 0.95, turns = 10

#### Prompt:

Problem:
……

A:

<code>

……

###BEGIN SOLUTION

[insert]

###END SOLUTION

…
</code>


A0: change example

A1: change logits(decimal places, array, etc)

A2: change output type (array -> dict, etc)

A3: analogy

A4: dimension(index) involved

A5: inverted operation

A6: order

A7: ±condition/operation

combinations involved, only show the highest level.

''',
'scipy':
'''
#### Setup
temperature = 0.7, topP = 0.95, 10 attempts.

#### Prompt:

Problem:

……

A:

<code>

……

###BEGIN SOLUTION

[insert]

###END SOLUTION

…

</code>

Origin: original question from stackoverflow(might be specified or simplified)

Function: Let model fill in a function.

A1: paraphrasing, seems not effective to Codex.

A2: change example

A3: analogy(min->max, column->row, etc)

A6: result type constraint.

A7: ±condition/operation

'''
}

hyper_links = {'numpy':'https://docs.google.com/document/d/1WjMXfe-zV5VvKfbUnyxauTBciPB1Bp82baaIrG3XffM/edit#',
'scipy': 'https://docs.google.com/document/d/1u_rGiLrLbH9Ac_OueTbmDFyLlWOtB0U56Ertp8ggW1Q/edit'}

st.write(intro[library])

st.write('If the demo seems a little confusing, feel free to check the document.', hyper_links[library])

dataset = []
#dataset = load_data(question_path)
with open(question_path) as f:
    lines = f.readlines()
    question = ''
    temp = {}
    tag = ''
    for line in lines:
        if line == 'Origin:\n' or line == 'Function:\n' or re.match(r'A\d:\n', line):
            if not tag:
                tag = line[:-2]
            else:
                temp[tag] = question
                question = ''
                tag = line[:-2]
        elif re.match(r'\d*\.\n', line):
            if tag:
                temp[tag] = question
                dataset.append(temp)
                question = ''
                tag = ''
                temp = {}
        else:
            if tag:
                question += line + '\n'

temp[tag] = question
dataset.append(temp)

# Select index
number = st.number_input("Insert a index: range from", 
    min_value = 0, max_value = len(dataset) - 1)
st.write('The current index is ', number)
data_index = int(number)

# Selece modification
options = tuple(dataset[data_index].keys())
modification = st.radio('Modification:',
            options = options
)


st.write(dataset[data_index][modification])