For the processed version, the exact details can be found in
the code of make_ames but a summary of the differences between
these data sets and ames_raw is:
All factors are unordered.
PID and Order are removed.
Spaces and special characters in column names where changed
to snake case. To be consistent, SalePrice was changed to
Sale_Price.
Many factor levels were changed to be more understandable
(e.g. Split_or_Multilevel instead of 080)
Many missing values were reset. For example, if the variable
Bsmt_Qual was missing, this implies that there is no basement
on the property. Instead of a missing value, the value of
Bsmt_Qual was changed to No_Basement. Similarly, numeric
data pertaining to basements were set to zero where appropriate
such as variables Bsmt_Full_Bath and Total_Bsmt_SF.
Garage_Yr_Blt contained many missing data and was removed.
Approximate longitude and latitude are included for the
properties. Also, note that there are 6 properties with
identical geotags. These are units within the same building.
For some properties, updated versions of the PID identifiers
were found and are replaced with new values.
make_ordinal_ames is the same as make_ames but many factor
variables were changed to class ordered (see below).
The documentation for ames_raw() contains descriptions of
the columns although, as noted above, the column names in
ames_raw() are slightly different from the processed
versions.
make_ames_new() creates a data set of new properties. These were populated
using less data sources than the original and lack a number of the condition
and quality. Both properties were unsold at the time of this writing.