Pre-processing the data

Fields visitorid and itemid are already numeric, but we still need to convert the events into numeric values.

1. We convert view events to 1, addtocart events to 2, and transaction events to 3 with the following code:

events.event.replace(to_replace=dict(view=1, 
                                     addtocart=2, 
                                     transaction=3), 
                     inplace=True)

2. Drop the transcationid and timestamp columns that we don't need:

events.drop(['transactionid'],axis=1,inplace=True)
events.drop(['timestamp'],axis=1,inplace=True)

3. Shuffle the dataset to get random data for training and test datasets:

events = events.reindex(np.random.permutation(events.index))

The dataset can also be shuffled with the following command:

events = events.sample(frac=1).reset_index(drop=True)

4. Split the data in train, valid, and test sets, as follows:

split_1 = int(0.8 * len(events))
split_2 = int(0.9 * len(events))
train = events[:split_1]
valid = events[split_1:split_2...