import math
import tangent
import numpy as np
import pandas as pd
import tensorflow as tf
from edward.models import Normal

def proforma(far, weighted_rent, parcel_size, land_cost,
             parking_rate, sqft_per_rate, sqft_per_stall, per_sqft_parking_cost,
             loan_to_cost_ratio, drawdown_factor, interest_rate, loan_fees, cap_rate,
             building_efficiency, profit_factor, building_cost_per_sqft, construction_months):
    """
    Pure-python implementation of proforma logic.
    """
    # Building bulk
    building_sqft = parcel_size * far

    # Parking
    parking_stalls = building_sqft * parking_rate / sqft_per_rate
    park_sqft = parking_stalls * sqft_per_stall
    park_cost = park_sqft * per_sqft_parking_cost

    # Total space
    total_built_sqft = building_sqft + park_sqft
    parking_sqft_ratio = park_sqft / total_built_sqft

    # Construction costs
    build_cost = building_cost_per_sqft * building_sqft
    cost = build_cost + park_cost
    cost_sqft = (cost / total_built_sqft) * profit_factor
    building_cost = building_sqft * cost_sqft # cost to build the new building
    total_construction_costs = building_cost + land_cost # add cost to buy current building

    # Financing costs
    loan_amount = total_construction_costs * loan_to_cost_ratio
    interest = loan_amount * drawdown_factor * (interest_rate / 12 * construction_months)
    points = loan_amount * loan_fees
    total_financing_costs = interest + points
    total_development_costs = total_construction_costs + total_financing_costs

    # Revenue
    usable_space = building_sqft * (1 - parking_sqft_ratio) * building_efficiency
    building_revenue = usable_space * weighted_rent / cap_rate

    # Profit
    profit = building_revenue - total_development_costs

    return profit

with tf.name_scope("bayesian_proforma"):
    """
    Directed graphical model of proforma.
    """
    weighted_rent = tf.constant(40.0, name='weighted_rent')
    far = tf.constant(4.0, name='far')
    parcel_size = tf.constant(30000.0, name='parcel_size')
    land_cost = tf.constant(3000000.0, name='land_cost')

    loan_to_cost_ratio = Normal(loc=tf.constant(0.7), scale=tf.constant(.3), name='loan_to_cost_ratio')
    drawdown_factor = Normal(loc=tf.constant(0.6), scale=tf.constant(.2), name='drawdown_factor')
    interest_rate = Normal(loc=tf.constant(0.05), scale=tf.constant(.02), name='interest_rate')
    loan_fees = Normal(loc=tf.constant(0.02), scale=tf.constant(0.005), name='loan_fees')
    building_efficiency = Normal(loc=tf.constant(0.7), scale=tf.constant(.3), name='building_efficiency')
    cap_rate = Normal(loc=tf.constant(0.05), scale=tf.constant(.02), name='cap_rate')
    profit_factor = Normal(loc=tf.constant(1.1), scale=tf.constant(.5), name='profit_factor')

    parking_rate = Normal(loc=tf.constant(1.0), scale=tf.constant(.4), name='parking_rate')
    sqft_per_rate = Normal(loc=tf.constant(1000.0), scale=tf.constant(200.0), name='sqft_per_rate')
    sqft_per_stall = Normal(loc=tf.constant(250.0), scale=tf.constant(70.0), name='sqft_per_stall')
    per_sqft_parking_cost = Normal(loc=tf.constant(110.0), scale=tf.constant(40.0), name='per_sqft_parking_cost')

    building_cost_per_sqft = Normal(loc=tf.constant(210.0), scale=tf.constant(70.0), name='building_cost_per_sqft')
    construction_months = Normal(loc=tf.constant(18.0), scale=tf.constant(6.0), name='construction_months')

    profoma_calcs = proforma(far, weighted_rent, parcel_size, land_cost,
             parking_rate, sqft_per_rate, sqft_per_stall, per_sqft_parking_cost,
             loan_to_cost_ratio, drawdown_factor, interest_rate, loan_fees, cap_rate,
             building_efficiency, profit_factor, building_cost_per_sqft, construction_months)

    profit = Normal(loc=profoma_calcs, scale=0.1, name='profit')


if __name__ == "__main__":

    far = 4.0
    loan_to_cost_ratio = .7
    drawdown_factor = 0.6
    interest_rate = 0.05
    loan_fees = 0.02
    building_efficiency = 0.7
    cap_rate = 0.05
    profit_factor = 1.1

    # Parking-type specific
    parking_rate = 1.0
    sqft_per_rate = 1000.0
    sqft_per_stall = 250.0
    per_sqft_parking_cost = 110

    # Parcel-specific  info
    weighted_rent = 40.0
    parcel_size = 30000.0
    land_cost = 3000000.0

    # Lookups
    building_cost_per_sqft = 210 ## This is a lookup (cost_per_sqft_for_this_height)
    construction_months = 18 ## A lookup of self.construction_months based on total_sqft

    ## Calculate profitability of a known example and confirm that result matches expectations
    result = proforma(far, weighted_rent, parcel_size, land_cost,
                     parking_rate, sqft_per_rate, sqft_per_stall, per_sqft_parking_cost,
                     loan_to_cost_ratio, drawdown_factor, interest_rate, loan_fees, cap_rate,
                     building_efficiency, profit_factor, building_cost_per_sqft, construction_months)
    print(result)
    assert result == 24402359.999999996

    dproforma = tangent.grad(proforma, wrt=range(17))

    ## Calculate the gradient of the proforma via auto-differentiation
    gradient = dproforma(far, weighted_rent, parcel_size, land_cost,
             parking_rate, sqft_per_rate, sqft_per_stall, per_sqft_parking_cost,
             loan_to_cost_ratio, drawdown_factor, interest_rate, loan_fees, cap_rate,
             building_efficiency, profit_factor, building_cost_per_sqft, construction_months)

    gradient = pd.Series(gradient)

    print(gradient)