Correcting Right Skew Data
In the slides we showed how you might use log transforms to fix positively skewed data (data whose distribution is mostly to the left). To correct negative skew (data mostly to the right) you need to take an extra step called "reflecting" before you can apply the inverse of \(\log\), written as (1/\(\log\)) to make the data look more like normal a normal distribution. Reflecting data uses the following formula to reflect each value: \((x_{\text{max}} +1) – x\).
This exercise is part of the course
Feature Engineering with PySpark
Exercise instructions
- Use the aggregate function
skewness()to verify that'YEARBUILT'has negative skew. - Use the
withColumn()to create a new column'Reflect_YearBuilt'and reflect the values of'YEARBUILT'. - Using
'Reflect_YearBuilt'column, create another column'adj_yearbuilt'by taking 1/log() of the values.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
from pyspark.sql.functions import log
# Compute the skewness
print(df.____({____: ____}).____())
# Calculate the max year
max_year = df.____({____: ____}).____()[0][0]
# Create a new column of reflected data
df = df.____(____, (max_year + 1) - df[____])
# Create a new column based reflected data
df = df.____(____, 1 / ____(df[____]))