ML Explainability App

hmeq_app.py

import streamlit as st
import pandas as pd 
import numpy as np
from PIL import Image
from swat import CAS, options
import json
import time
import math
import SessionState
import altair as alt
import random
import io 
import plotly.express as px
import base64


state = SessionState.get(swat_sess=None,samplepd=None,scored=None, numtopics =None)

def get_table_download_link(df):
    """Generates a link allowing the data in a given panda dataframe to be downloaded
    in:  dataframe
    out: href string
    """
    csv = df.to_csv(index=False)
    b64 = base64.b64encode(
        csv.encode()
    ).decode()  
    return f'<a href="data:file/csv;base64,{b64}" download="sentiment.csv">Download Sentiment scores file</a>'

def get_topic_download_link(df):
    """Generates a link allowing the data in a given panda dataframe to be downloaded
    in:  dataframe
    out: href string
    """
    csv = df.to_csv(index=False)
    b64 = base64.b64encode(
        csv.encode()
    ).decode()  # some strings <-> bytes conversions necessary here
    return f'<a href="data:file/csv;base64,{b64}" download="sentiment.csv">Download Topic scores file</a>'

image = Image.open('images/sasLogo.png')
def dicttopd(datadictionary):
    for key in datadictionary:
        datadictionary[key] = [datadictionary[key]]
    return pd.DataFrame.from_dict(datadictionary)


st.markdown(
    """
<style>

.sidebar .sidebar-content {
    background-image: linear-gradient(#2e7bcf,#2e7bcf);
    color: white;
}
.sidebar-content .Widget>label {
    color: white;
}
.Widget>label {
    color: black;
}
[class^="st-b"]  {
    color: black;
}
.st-bb {
    background-color: transparent;
}
.st-at {
    background-color: white;
}

.reportview-container .main footer, .reportview-container .main footer a {
    color: white;
}
header .decoration {
    background-image: none;
}

</style>
""",
    unsafe_allow_html=True,
)
st.sidebar.image(image)
st.sidebar.write("""
# Explainable AI App
""")
hostname = st.sidebar.text_input('Please Enter host URI')

username= st.sidebar.text_input('Please Enter host Username')

password= st.sidebar.text_input('Please Enter host Password',type="password")
def logon(hostname, username, password):
    s = CAS(hostname=hostname, protocol='cas', 
            username=username, password=password)
    s.loadActionSet('autotune')
    s.loadactionset('aStore')
    s.loadactionset('table')
    s.loadactionset('decisionTree')
    s.loadactionset("explainModel")
    s.loadactionset('table')
    s.loadActionSet(actionSet="sentimentAnalysis")     
    s.loadActionSet(actionSet="textMining")     
    return s
s = None 
if st.sidebar.button('Login'):
    try:
        s = logon(username, password)
        state.swat_sess= s

    except: 
        st.sidebar.write('Please Enter a valid user name and password')

demo_title = st.sidebar.text_input('Please Enter Demo Title')
heading = """
    # {} 
""".format(demo_title)

option = st.sidebar.selectbox(
    'Select the scenario you wish to demonstrate',
    ('Home Loan Default','Concrete Strength Predictor','Telecom Customer Churn','Financial Fraud Detection','Retail Price Forecast','Predict Early Stage Diabetes Risk','Bank Customer Churn','Heart Disease Prediction','NLP : Sentiment','NLP : Sentiment and SVD Topics'))

st.sidebar.button('Download App Code')
st.sidebar.button('Download App Docker Image')
if option !='Financial Fraud Detection':
    st.write(heading)
if option == 'Home Loan Default':
    
    st.write('Please choose variable inputs: ')
    def user_input_features():
        LOAN = st.slider('Loan Amount', 1000, 90000, 20000)
        MORTDUE = st.slider('Mortgage Due', 2000, 400000, 10000)
        VALUE = st.slider('Value of Property', 6000, 860000, 75000)
        YOJ = st.slider('Years on the Job', 0, 45, 20)
        DEROG	 = st.slider('Number of derogatory remarks', 0, 10, 4)
        DELINQ = st.slider('Number of delinquencies', 0, 16, 8)
        CLAGE	 = st.slider('Age of credit line in months', 0, 240, 12)
        NINQ	 = st.slider('Number of inquiries', 0, 20, 10)
        CLNO = st.slider('Number of credit lines', 0, 75, 35)
        DEBTINC = st.slider('Debt to Income ratio', 0.5, 205.0, 80.0)
        JOB = st.selectbox('Job/Occupation',
                        ('Mgr', 'Office', 'Other', 'ProfExe', 'Sales', 'Self'))
        REASON = st.selectbox('Reason',
                        ('DebtCon', 'HomeImp'))


        data = {'LOAN': LOAN,
                'MORTDUE': MORTDUE,
                'VALUE': VALUE,
                'YOJ': YOJ,
                'DEROG':DEROG,
                'DELINQ': DELINQ,
                'CLAGE': CLAGE,
                'NINQ': NINQ,
                'CLNO': CLNO,
                'DEBTINC': DEBTINC,
                'JOB': JOB,
                'REASON': REASON,
                }
        features = pd.DataFrame(data, index=[0])
        return features
   
    samplepd = user_input_features()
    state.samplepd = samplepd
    
    st.write('Please verify that the input data is correct: ')
    st.write(samplepd)
    try:
        s = state.swat_sess
    except:
        st.write('Please log in to the SAS Viya to use the AI Composer')
    @st.cache
    def score(samplepd):
        s.upload(samplepd,casout={'name' : 'realtime', 'caslib' : 'public','replace' : True})
        s.aStore.score(rstore = {"caslib":"public","name":"hmeqTestAstore"},
                            table = {"caslib":'public',"name":'realtime'},
                            out = {"caslib":'public',"name":'realscore', 'replace':True})
        scoredData = s.CASTable(name='realscore',caslib='public')
        datasetDict = scoredData.to_dict()
        scores = pd.DataFrame(datasetDict, index=[0])
        return scores
    @st.cache
    def explainML(samplepd):
        s.upload(samplepd,casout={'name' : 'realtime', 'caslib' : 'public','replace' : True})
        shapvals = s.linearExplainer(
                    table           = {"name" : 'hmeqTest','caslib':'public'},
                    query           = {"name" : 'realtime','caslib':'public'},
                    modelTable      = {"name" :"hmeqTestAstore",'caslib':'public'},
                    modelTableType  = "ASTORE",
                    predictedTarget = 'P_BAD1',
                    seed            = 1234,
                    preset          = "KERNELSHAP",
                    inputs          = ['LOAN','MORTDUE','VALUE','YOJ','DEROG','DELINQ','CLAGE','NINQ','CLNO','DEBTINC','REASON', 'JOB','BAD'],
                    nominals        = ['REASON', 'JOB','BAD']
                    )
        shap1 = shapvals['ParameterEstimates']
        shap = shap1[['Variable','Estimate']][1:10]
        labels = list(shap['Variable'])
        data = list(shap['Estimate'])
        target = 'BAD'
        labels_n = []
        data_n  = []
        i = 0
        for i in range(len(labels)):
            if labels[i] != target:
                labels_n.append(labels[i])
                data_n.append(data[i])

        return labels_n, data_n

    if st.button('Score'):
        try:
            scores = score(samplepd)
            st.write('The predicted outcome is: ')
            st.write(scores)
            samplepd = state.samplepd
            samplepd['BAD'] = int(scores['I_BAD'][0].strip(" "))
            state.scored = samplepd

        except: 
            st.write("Are you sure you're signed into the SAS server?")

    if st.button('Explain my prediction'):
        
        samplepd = state.scored
        st.write('The predicted value for target {} is: {}, and the contributions of the variables are as follows:'.format('BAD',samplepd['BAD'][0]))
        labels, data = explainML(samplepd)
        source = pd.DataFrame({
            'Predictors': labels,
            'Estimated Impact': data
        })

        c = alt.Chart(source).mark_bar().encode(
            x='Predictors',
            y='Estimated Impact',
            color=alt.Color('Predictors'), tooltip=['Predictors', 'Estimated Impact']
        )
        st.altair_chart(c, use_container_width=True)