# -*- coding: utf-8 -*-
"""Untitled32.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1fKN0jOoDSOaUCMAAoxNUnSUd-HPcRXNZ
"""
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from lightgbm import LGBMClassifier
from sklearn.model_selection import train_test_split

import streamlit as st
import pandas as pd

import numpy as np

data = pd.read_csv('archive (8).zip')
data.head()

to_use_cols = ['as_of_year', 'county_name', 'applicant_sex_name', 'action_taken_name', 'loan_amount_000s', 'applicant_income_000s', 'state_name', 'property_type_name', 'loan_type_name']

data_reduced = data[to_use_cols]

data_no_na = data_reduced.dropna()


succeeded = ['Loan originated', 'Loan purchased by the institution', ]
failed = ['Application approved but not accepted', 'Preapproval request denied by financial institution', 'Application denied by financial institution', 'Preapproval request approved but not accepted']
user_error = ['File closed for incompleteness', 'Application withdrawn by applicant', ]

mapped = {tuple(succeeded): 1, tuple(failed): 2, tuple(user_error): 3}

def mapped(x):
  if x in succeeded:
    return 1
  else:
    return 0

data_no_na.action_taken_name = data_no_na.action_taken_name.apply(mapped)


mapped_type = {
    'Conventional': 0,
    'FHA-insured': 1,
    'VA-guaranteed':2,
    'FSA/RHS-guaranteed':3
}

data_no_na.loan_type_name.apply(lambda x: mapped_type[x])

data_no_na['loan_encoded'] = data_no_na.loan_type_name.apply(lambda x: mapped_type[x])

data_no_na.property_type_name.value_counts()

data_no_na['property_encoded'] = data_no_na.property_type_name.apply(lambda x: 1 if x == 'Manufactured housing' else 0)

data_no_na.state_name.value_counts()

data_no_na.county_name.value_counts()


code = {}
i = 0
for county in data_no_na.county_name.unique():
  code[county] = i
  i += 1

data_no_na['county_code']= data_no_na.county_name.map(code)

data_no_na.head(2)

data_no_na['sex_encoded'] = data_no_na.applicant_sex_name.apply(lambda x: 1 if x == 'Male' else 0)

data_no_na.head(2)

cols = ['county_code', 'sex_encoded', 'property_encoded', 'loan_encoded', 'applicant_income_000s', 'loan_amount_000s', 'action_taken_name']

train = data_no_na[cols]

X_train, X_test, y_train,  y_test = train_test_split(train.drop('action_taken_name', axis=1), train.action_taken_name, test_size=0.3, random_state=42)


gbm = LGBMClassifier(n_estimators=200)
random = RandomForestClassifier(n_estimators=200)
tree = DecisionTreeClassifier()

gbm.fit(X_train, y_train)
random.fit(X_train, y_train)
tree.fit(X_train, y_train)

def mapping(**kwargs):
  kwargs['county'] = code[kwargs['county']]
  kwargs['sex'] = 1 if kwargs['sex'] == 'Male' else 0
  kwargs['property_type'] =  1 if kwargs['property_type'] == 'Manufactured housing' else 0
  kwargs['loan'] = mapped_type[kwargs['loan']]
  kwargs['income'] = float(kwargs['income'])
  kwargs['loan_amount'] = float(kwargs['loan_amount'])
  return kwargs


st.title('Loan Approval for Washington House')
st.dataframe(data_no_na.head())


col1, col2 = st.columns([3, 1])
with col1:
  st.text('Please, fill this form')

  with st.form("my_form"):
    county = st.selectbox('County', data_no_na.county_name.unique().tolist())
    sex = st.selectbox('Sex', ['Male', 'Female'])
    property_type = st.selectbox('Property Type', data_no_na.property_type_name.unique().tolist())
    loan = st.selectbox('Loan Type', data_no_na.loan_type_name.unique().tolist())
    income = st.number_input('Your Yearly income (in 000$)')
    loan_amount = st.number_input('Loan Amount')

    model_choice = col2.selectbox('Choose model', ['RandomForest', 'Tree', 'LGBM'])
    # Every form must have a submit button.
    submitted = st.form_submit_button("Submit")
    if submitted:
        col2.info('Predicting')
        new_demand = np.array([list(mapping(county=county, sex=sex, property_type=property_type, loan=loan, income=income, loan_amount=loan_amount).values())])
        if model_choice == 'Tree':
          result = tree.predict(new_demand)
        elif model_choice == 'RandomForest':
          result = random.predict(new_demand)
        else:
          result = gbm.predict(new_demand)

        if result[0]:
            col2.success('Accepted')
        else:
            col2.error("Rejected")