import streamlit as st
import pickle
import pandas as pd
import torch
import numpy as np

cosine_scores = pickle.load(open('cosine_scores.pkl','rb'))
coursedf = pd.read_pickle('course_df.pkl')          # course_df uses titles to generate course recommendations
course_df_new = pd.read_pickle('course_df_new.pkl') #course_df_new makes recommendations using the entire description

course_title_list = [i + ": " + j for i, j in zip(coursedf['ref'].to_list(), coursedf['title'].to_list())]

def get_random_course():
    row=coursedf.sample(1)
    return row['ref'], row['title'] 

def recommend(index):
    pairs = {}

    for i in range(len(coursedf)):
        pairs[coursedf.iloc[i,1]]=cosine_scores[index][i]

    sorttemp = sorted(pairs.items(), key=lambda x:x[1], reverse=True)
    sorted_final = dict(sorttemp[1:31])

    return list(sorted_final.keys())

st.set_page_config(page_title='DiscoverCourses', page_icon=':bird:')
st.header('DiscoverCourses')
st.write('')

selected_course = st.selectbox('Pick a course from the dropdown:',course_title_list)

container = st.container()
maincol1, maincol2 = container.columns(2)
st.write('')

if maincol1.button('Recommend by title',use_container_width=True):
    output=recommend(np.where((coursedf['ref']+": "+coursedf['title']) == selected_course)[0][0])
    for result in output:
        index=np.where(coursedf['title'] == result)[0][0]
        course_id=coursedf.iloc[index,0]
        st.subheader(course_id+": "+result)
        with st.expander("See description"):
            st.write(course_df_new.iloc[index,3]) #Using the new coursedf because it has proper descriptions for each course
        link = "[ExploreCourses](https://explorecourses.stanford.edu/search?q="+course_id+"+"+result.replace(" ","+")+")"
        st.markdown(link, unsafe_allow_html=True)
        link = "[Carta](https://carta-beta.stanford.edu/results/"+course_id+")"
        st.markdown(link, unsafe_allow_html=True)
        st.divider()
        
if maincol2.button('Recommend by description',use_container_width=True):
    index_new=np.where((coursedf['ref']+": "+coursedf['title']) == selected_course)[0][0]
    rec_list=course_df_new.iloc[index_new,2]
    for result in rec_list:
        index=np.where(coursedf['title'] == result)[0][0]
        course_id=coursedf.iloc[index,0]
        st.subheader(course_id+": "+result)
        with st.expander("See description"):
            st.write(course_df_new.iloc[index,3]) #Using the new coursedf because it has proper descriptions for each course
        link = "[ExploreCourses](https://explorecourses.stanford.edu/search?q="+course_id+"+"+result.replace(" ","+")+")"
        st.markdown(link, unsafe_allow_html=True)
        link = "[Carta](https://carta-beta.stanford.edu/results/"+course_id+")"
        st.markdown(link, unsafe_allow_html=True)
        st.divider()

st.write('© 2023 Rushank Goyal. All rights reserved. Source for the all-MiniLM-L6-v2 model: Wang, Wenhui, et al. "MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers." arXiv, 25 Feb. 2020, doi:10.48550/arXiv.2002.10957.')