app.py

import streamlit as st
import transformers as tf
import plotly.graph_objects as go
import matplotlib.cm as cm
import pandas as pd


# Function to load and cache models
@st.experimental_singleton(show_spinner=False)
def load_model(username, prefix, model_name):
    p = tf.pipeline('text-classification', f'{username}/{prefix}-{model_name}')
    return p

@st.experimental_singleton(show_spinner=False)
def load_pickle(f):
    return pd.read_pickle(f)

def get_results(model, c):
    res = model(c)[0]
    label = float(res['label'].split('_')[1])
    score = res['score']
    return {'label': label, 'score': score}

def run_models(model_names, models, c):
    results = {}
    for mn in model_names:
        results[mn] = get_results(models[mn], c)
    return results


st.title('Assess the *QuAL*ity of your feedback')
st.caption(
"""Medical education *requires* high-quality feedback, but evaluating feedback
is difficult and time-consuming. This tool uses NLP/ML to predict a validated
feedback quality metric known as the QuAL Score. *Try it for yourself!*
""")

### Load models
# Specify which models to load 
USERNAME = 'maxspad'
PREFIX = 'nlp-qual'
models_to_load = ['qual', 'q1', 'q2i', 'q3i']
n_models = float(len(models_to_load))
models = {}
# Show a progress bar while models are downloading, 
# then hide it when done
lc_placeholder = st.empty()
loader_container = lc_placeholder.container()
loader_container.caption('Loading models... please wait...')
pbar = loader_container.progress(0.0)
for i, mn in enumerate(models_to_load):
    pbar.progress((i+1.0) / n_models)
    models[mn] = load_model(USERNAME, PREFIX, mn)
lc_placeholder.empty()

### Load example data
examples = load_pickle('test.pkl')

### Process input
ex = examples['comment'].sample(1).tolist()[0]
try:
    ex = ex.strip().replace('_x000D_', '').replace('nan', 'blank')
except:
    ex = 'blank'
if 'comment' not in st.session_state:
    st.session_state['comment'] = ex
with st.form('comment_form'):
    comment = st.text_area('Try a comment:', value=st.session_state['comment'])
    left_col, right_col = st.columns([1,9], gap='medium')
    submitted = left_col.form_submit_button('Submit')
    trying_example = right_col.form_submit_button('Try an example!')
    
    if submitted:
        st.session_state['button_clicked'] = 'submit'
        st.session_state['comment'] = comment
        st.experimental_rerun()
    elif trying_example:
        st.session_state['button_clicked'] = 'example'
        st.session_state['comment'] = ex
        st.experimental_rerun()
    
results = run_models(models_to_load, models, st.session_state['comment'])

tab_titles = ['Overview', 'Q1 - Level of Detail', 'Q2 - Suggestion Given', 'Q3 - Suggestion Linked', 'About']
tabs = st.tabs(tab_titles)

with tabs[0]:
    cmap = cm.get_cmap('RdYlGn')
    color = cmap(results['qual']['label'] / 6.0)
    color = f'rgba({int(color[0]*256)}, {int(color[1]*256)}, {int(color[2]*256)}, {int(color[3]*256)})'

    fig = go.Figure(go.Indicator(
        domain = {'x': [0, 1], 'y': [0, 1]},
        value = results['qual']['label'],
        mode = "gauge+number",
        title = {'text': "QuAL"},
        gauge = {'axis': {'range': [None, 5]},
                'bgcolor': 'lightgray',
                'bar': {'color': color, 'thickness': 1.0},
                
                }
        ), layout=go.Layout(margin=dict(t=0, b=135)))#, layout=go.Layout(width=750, height=300))# layout={'paper_bgcolor': 'rgb(245,245,245)'})#,

    cols = st.columns([7, 3])
    with cols[0]:
        st.plotly_chart(fig, use_container_width=True)
    with cols[1]:

    
        # cols = st.columns(3)
        # cols[0].markdown('#### Level of Detail')
        q1lab = results['q1']['label']
        if q1lab == 0:
            md_str = '😥 None'
        elif q1lab == 1:
            md_str = '😐 Low'
        elif q1lab == 2:
            md_str = '😊 Medium'
        elif q1lab == 3:
            md_str = '😁 High'
        # cols[0].markdown(md_str)
        cols[1].metric('Level of Detail', md_str, 
            help='How specific was the evaluator in describing the behavior?')

        q2lab = results['q2i']['label']
        if q2lab == 0:
            md_str = '✅ Yes'
        else:
            md_str = '❌ No'
        cols[1].metric('Suggestion Given', (md_str),
            help='Did the evaluator give a suggestion for improvement?')

        q3lab = results['q3i']['label']
        if q3lab == 0:
            md_str = '✅ Yes'
        else:
            md_str = '❌ No'
        cols[1].metric('Suggestion Linked', md_str,
            help='Is the suggestion for improvement linked to the described behavior?')