Introduction

In this assignment you will build the best model possible to predict $$SalePrice$.$

Late Submissions

Late submissions are permitted but $10\%$ will be deducted from the assignment for each day

that the assignment is late. As always, sharing your code with others and copying code is not

permitted.

Deliverables

$$ In your zipped deliverable folder please submit only three files:

$1.$ Your report $($please use word and do not submit a pdf document or marks will be

deducted$).($firstname$\_$report.docx$)$

$2.$ Your modelling code for your best model which runs from start to finish without

error. $($firstname$\_$test$\_$train.py$)$

$3.$ Your production code $($firstname$\_$production.py$).$ The production code must:

$$ Prepare the data, use your pre$-$trained top model to make predictions

and output the predictions to a csv file.

$$ Contain all data preparation code.

$$ Not delete any data.

$$ Not contain any model training or testing code.

$$ Run without error.

$$ Load data from a file that uses the exact format as the

asgn$1\_$houseprices$\_$mystery.csv file. The number of rows may vary.

$$ Output to your predictions in the exact same format as shown with the

sample asgn$1\_$houseprices$\_$predictions.csv file.

$($Please see the hints for this assignment in the assignment folder for an example of

production code$).$

Marks will be deducted if deliverables are not submitted in the format requested.

Competition

$$ The person in the class who gets the lowest RMSE when I test their score code without

error will get $10\%$ added to their assignment mark.

$$ The next lowest RMSE $($without error$)$ will get a $5\%$ bonus added to their assignment

mark.

$$ All participants who make an honest effort earn good karma.

Disqualification from the competition will result if too many insignificant variables exist in the

score code. Excessive overfitting is not allowed. Late submissions are not eligible for the bonus.

You must use linear regression. You cannot use alternative algorithms.

Report

Please keep your report to limited between $8$ to $12$ pages. Be creative here about how you fit

the content in it and what you fit in$.$ Your employer is not actually familiar with the attributes in

the dataset so at least report on the numbers but you do not have to be an expert in the

domain to be a good analyst. Initiative, good self$-$judgment and professional delivery are

appreciated by your employer. Aim for efficient reader$-$friendliness with comparison tables and

uncluttered visualizations where appropriate rather than large dumps of data in the report.

Not Evident

$0$ points

Little to none

$6$ points

Needs Improvement

$14$ points

About right; as expected

$19$ points

Above average

$23$ points

Extraordinary $-$ Well above the expected norm

$25$ points

Report Introduction $(2$ marks$)$

o Describe the problem that you are attempting to solve. Mention the best features that

were found at the start of the report.

Exploratory Data Analysis $(23$ marks$)$

o For the EDA section please assume you are presenting an overview of the data to a not$-$

very$-$technical group of managers. Please use good sense when assembling the EDA. Be

mindful of your time and also of the time for the group you are serving. In your report,

please be sure to focus on the variables that make a difference in your model right

away. Do not spend too much time discussing variables that are not relevant.

o Prepare a summary of the data.

o Show the correlations between the target and at least all predictor variables which are

in the final best model.

o Highlight features of interest and how they might impact the predictions positively or

negatively.

o You may show scatter plots, histograms, or other plots for relevant variables where

appropriate.

o Create a visual summary $($s$)$ which categorizes the target range into three or four

groups. For example; $$Group A has xx traits and high numbers of yy$.$ Group B has ww

traits and moderate amounts of zz$.$ Your employer is not really sure what is needed here but has asked you to figure out how to present this information in an easy$-$to$-$read

format. Please see the hints folder for suggestions.

Development $(15$ marks$)$

Not Evident

$0$ points

Little to none

$3$ points

Needs Improvement

$7$ points

About right; as expected

$11$ points

Above average

$13\mathrm{.}5$ points

Extraordinary $-$ Well above the expected norm

$15$ points

You are to make at least three models:

o One of the models must include binned and dummy variables. Take a reasonable effort

to find binned and dummy values which ideally boost performance $($it may not though$).$

Please be creative here.

o Remember categorical variables are good candidates for dummy variables and possibly

binning.

o One model must include outlier treatment of some kind. Please be creative here.

o One model must not include binned and dummy variables.

o Experimentation with many different variable combinations is encouraged and

necessary to discover the top performing models.

o Uses cross fold validation with truly random data $($in other words remove random$\_$state

from your code$)$

Model Eval

Incl