bciavm 1.21.13

bciAVM is a machine learning pipeline used to predict property prices.

The Blockchain & Climate Institute (BCI) is a progressive think tank providing leading expertise in the deployment of emerging technologies for climate and sustainability actions.

As an international network of scientific and technological experts, BCI is at the forefront of innovative efforts, enabling technology transfers, to create a sustainable and clean global future.

Automated Valuation Model (AVM)

About

AVM is a term for a service that uses mathematical modeling combined with databases of existing properties and transactions to calculate real estate values. The majority of automated valuation models (AVMs) compare the values of similar properties at the same point in time. Many appraisers, and even Wall Street institutions, use this type of model to value residential properties. (see What is an AVM Investopedia.com)

For more detailed info about the AVM, please read the About paper in the resources directory.

Valuation Process

Key Functionality

• Supervised algorithms
• Tree-based & deep learning algorithms
• Feature engineering derived from small clusters of similar properties
• Ensemble (value blending) approaches

Set the required AWS Environment Variables

export ACCESS_KEY=YOURACCESS_KEY
export SECRET_KEY=YOURSECRET_KEY
export BUCKET_NAME=bci-transition-risk-data
export TABLE_DIRECTORY=/dbfs/FileStore/tables/

Install from PyPI

pip install bciavm

Start

Load the training data from the BCI S3 bucket

from bciavm.core.config import your_bucket
from bciavm.utils.bci_utils import ReadParquetFile, get_postcodeOutcode_from_postcode, get_postcodeArea_from_outcode, drop_outliers, preprocess_data
import pandas as pd

dfPricesEpc = pd.DataFrame()
dfPrices = pd.DataFrame()

yearArray = ['2020', '2019']
for year in yearArray:
    singlePriceEpcFile = pd.DataFrame(ReadParquetFile(your_bucket, 'epc_price_data/byDate/2021-02-04/parquet/' + year))
    dfPricesEpc = dfPricesEpc.append(singlePriceEpcFile)

dfPricesEpc['POSTCODE_OUTCODE'] = dfPricesEpc['Postcode'].apply(get_postcodeOutcode_from_postcode)
dfPricesEpc['POSTCODE_AREA'] = dfPricesEpc['POSTCODE_OUTCODE'].apply(get_postcodeArea_from_outcode)
dfPricesEpc.groupby('TypeOfMatching_m').count()['Postcode']

Preprocess & split the data for training/testing

import bciavm
X_train, X_test, y_train, y_test = bciavm.preprocess_data(dfPricesEpc)

Build the pipeline and get the default pipeline parameters

from bciavm.pipelines import RegressionPipeline

class AVMPipeline(RegressionPipeline):
        custom_name = 'AVM Pipeline'
        component_graph = {
            "Preprocess Transformer": ["Preprocess Transformer"],
            'Imputer': ['Imputer', "Preprocess Transformer"],
            'One Hot Encoder': ['One Hot Encoder', "Imputer"],
            'K Nearest Neighbors Regressor': ['K Nearest Neighbors Regressor', 'One Hot Encoder'],
            "XGBoost Regressor": ["XGBoost Regressor", 'One Hot Encoder'],
            'MultiLayer Perceptron Regressor': ['MultiLayer Perceptron Regressor',  'One Hot Encoder'],
            'Final Estimator': ['Linear Regressor', "XGBoost Regressor", 'MultiLayer Perceptron Regressor', 'K Nearest Neighbors Regressor']
        }

avm_pipeline = AVMPipeline(parameters={})
avm_pipeline.parameters

Fit the pipeline

avm_pipeline.fit(X_train, y_train)

Score the pipeline

avm_pipeline.score(  X_test, 
                     y_test, 
                     objectives=['MAPE',
                               'MdAPE',
                               'ExpVariance',
                               'MaxError',
                               'MedianAE',
                               'MSE',
                               'MAE',
                               'R2',
                               'Root Mean Squared Error'])

Next Steps

Read more about bciAVM on our documentation page:

How does it relate to BCI Risk Modeling?

Technical & financial support for AVM development was provided by GCODE.ai