Engineering features

Machine Learning for Finance in Python

Nathan George

Data Science Professor

nonlinear feature correlation

One problem with linear models

# add non-linear interaction term for a linear model
SMAxRSI = amd_df['14-day SMA'] * amd_df['14-day RSI']

Some models that don't require manually creating interaction features:

Decision-tree-based models

Random forests
Gradient boosting

Others

neural networks

volume vs price change

price and volume

Volume features

amd_df['Adj_Volume_1d_change'] = amd_df['Adj_Volume'].pct_change()

one_day_change = amd_df['Adj_Volume_1d_change'].values
amd_df['Adj_Volume_1d_change_SMA'] = talib.SMA(one_day_change,
                                                timeperiod=10)

Datetime feature engineering

graphic of datetime with arrows to day of week, month, year

Extracting the day of week

print(amd_df.index.dayofweek)

Int64Index([2, 3, 4, 0, 1, 2, 3, 4, 0, 1,
            ...
            1, 2, 3, 4, 0, 1, 2, 3, 4, 0],
           dtype='int64', name='Date', length=4807)

Dummies

days_of_week = pd.get_dummies(amd_df.index.dayofweek,
                                prefix='weekday',
                                drop_first=True)

print(days_of_week.head())

            weekday_1  weekday_2  weekday_3  weekday_4
Date
2018-04-10          1          0          0          0
2018-04-11          0          1          0          0
2018-04-12          0          0          1          0
2018-04-13          0          0          0          1
2018-04-16          0          0          0          0

correlation plot of new features

Engineer some features!

Machine Learning for Finance in Python