Bach and sklearn
With Objectiv you can do all your analysis and Machine Learning directly on the raw data in your SQL database. This example shows in the simplest way possible how you can use Objectiv to create a basic feature set and use sklearn to do machine learning on this data set. We also have an example that goes deeper into feature engineering here.
At first we have to install the open model hub and instantiate the Objectiv DataFrame object. See Getting started with Objectiv for more info on this.
This object points to all data in the data set. Without any aggregation, this dataset is too large to for pandas and sklearn. For the data set that we need, we aggregate to user level, at which point it is small enough to fit in memory.
We create a data set of per user all the root locations that the user clicked on.
# extract the root location from the location stack
df['root'] = df.location_stack.ls.get_from_context_with_type_series(type='RootLocationContext', key='id')
# only look at press events and count the root locations
features = df[(df.event_type=='PressEvent')].groupby('user_id').root.value_counts()
# unstack the series, to create a DataFrame with the number of clicks per root location as columns
features_unstacked = features.unstack(fill_value=0)
Now we have a basic feature set that is small enough to fit in memory. This can be used with sklearn, as we demonstrate in this example.
from sklearn import cluster
# export to pandas now
pdf = features_unstacked.to_pandas()
# do the clustering using the pandas DataFrame and set the labels as a column to that DataFrame
est = cluster.KMeans(n_clusters=3)
df['cluster'] = est.labels_
Now you can use the created clusters on your entire data set again if you add it back to your DataFrame. This is simple, as Bach and pandas are cooperating nicely. Your original Objectiv data now has a ‘cluster’ column.
kfeatures_unstacked['cluster'] = pdf['cluster']
df_with_cluster_results = df.merge(features_unstacked[['cluster']], on='user_id')
You can use this column, just as any other. For example you can now use your created clusters to group models from the model hub by:
# Expected output:
# 0 0 days 00:09:18.204353
# 1 0 days 00:10:25.104636
# 2 0 days 00:20:43.561232
# Name: session_duration, dtype: timedelta64[ns]