Hands-On Machine Learning for Algorithmic Trading

How to prepare the data

We load the Quandl adjusted stock price data (see the instructions on GitHub on how to obtain the source data) as follows (see the build_dataset notebook):

prices = (pd.read_hdf('../data/assets.h5', 'quandl/wiki/prices')          .adj_close          .unstack().loc['2007':])prices.info()DatetimeIndex: 2896 entries, 2007-01-01 to 2018-03-27Columns: 3199 entries, A to ZUMZ

We start by generating weekly returns for close to 2,500 stocks without missing data for the 2008-17 period, as follows:

returns = (prices           .resample('W')           .last()           .pct_change()           .loc['2008': '2017']           .dropna(axis=1)           .sort_index(ascending=False))returns.info()DatetimeIndex: 522 entries, 2017-12-31 to 2008-01-06Columns: 2489 entries, A to ZUMZ

We'll use 52-week sequences, ...