praxis-timeseries-client 0.0.4

A Darts Time Series client for facilitating model development and data exploration.

Praxis Time Series Client

Current version: 0.0.4 Python version: >= 3.7

A package for quick visualization of a target Darts time series objects alongside covariates and discrete data.

Usage

The main component of this package is the DartsInterface class found in praxis_timeseries_client.darts. This synthesizes together your Time Series objects under the same hood, so you can interact with them together in a single graph and run models in a plug-and-play manner with your covariates.

from praxis_timeseries_client.darts import DartsInterface
from darts import TimeSeries

interface = DartsInterface(
    target_ts=TimeSeries.from_dataframe(...),
    past_covariates=TimeSeries.from_dataframe(...),
    future_covariates=TimeSeries.from_dataframe(...),
    discrete_vars={...})

Now, you can plot with the interface:

interface.plot(components=[...]).show() # Plot all components, or an ordered subset

And run forecasts and backtests with a model pre-trained on the covariates:

# make sure ts_train == target_ts from above
model = LinearRegressionModel(lags=50, lags_future_covariates=(100, 50), output_chunk_length=30)
model.fit(series=ts_train, future_covariates=ts_future_covs)

# run a backtest and plot it
backtest = interface.backtest(
    model,
    start=pd.Timestamp(startDate),
    forecast_horizon=int(forecastHorizon),
    stride=int(stride))
backtest.run(past_covariates=None)
backtest.plot().show()

# or run a forecast and plot it
forecast = interface.forecast(model, n=50)
forecast.run(past_covariates=None)
forecast.plot().show()

Hopefully this is enough to get you started using the interfaces!