Feature Store for Timeseries

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Modern feature store

Our feature store is optimised for timeseries data

Pre-packaged data

We can pre-populate your feature store with ready-to-use timeseries

Production ready

Serve features to production ML models

What is it for?

Organise timeseries data

Feature stores provide a way to organise much of the data preparation required when building machine-learning/AI models. They allow data scientists to become more productive, and ease the path between research and production.
Read more about feature stores for timeseries

Designed for data scientists

We've focused on making our feature store easy-to-use with simple Python/REST interfaces. Get up and running with no complex software installation.
Code Editor
import bytehub as bh

fs = bh.feature_store()

df = fs.get_timeseries(
['weather.temperature', 'sales.volume'],
from_date='2020-01-01', to_date='2020-08-01',
freq='1d'
)

df.head()

Clean and simple

We've created a simple, easy-to-use interface that fits into your existing data-science workflow

Code Editor
print(df.value[0])

{
'forecast': [0.5, 1.2, 4.5, 1.6],
'forecast_type': 'hourly'
}

Flexible

ByteHub's feature store can deal with a wide-variety of timeseries, from simple values to complex arrays and dictionary structures

Code Editor
# Get the latest feature values
df = fs.get_last(
['feature_1', 'feature_2', 'feature_3']
)

result = model.predict(df)

Production-ready

Use the same API to feed the latest features into your production models