Today at @nixtlainc: fast and interpretable forecasting for multiple seasonalities. β²οΈπ
This is exciting news for data scientists in the Python ecosystem. π§΅
#reinventforecasting
This is exciting news for data scientists in the Python ecosystem. π§΅
#reinventforecasting
Time series are the operational DNA of the world. π§¬
Most business data (sales, demand, electricity load, sensors) is stored in tabular format with time indexes.
Most forecasting models in Python are only suited for single seasonal data. β οΈ
#reinventforecasting
Most business data (sales, demand, electricity load, sensors) is stored in tabular format with time indexes.
Most forecasting models in Python are only suited for single seasonal data. β οΈ
#reinventforecasting
However, many time series have multiple seasonalities.
Multiple seasonality is traditionally present in data sampled at a low frequency. π‘
For example, hourly electricity data exhibits daily and weekly seasonality. ππβοΈ
#reinventforecasting
Multiple seasonality is traditionally present in data sampled at a low frequency. π‘
For example, hourly electricity data exhibits daily and weekly seasonality. ππβοΈ
#reinventforecasting
Traditional models are only able to model one seasonality.
Multiple Seasonal-Trend decompositions with LOESS (MSTL) by @kbandara164, @robjhyndman, and
@CBergmeir was created for that specific challenge.
But until now, MSTL was only available in #RStats
#reinventforecasting
Multiple Seasonal-Trend decompositions with LOESS (MSTL) by @kbandara164, @robjhyndman, and
@CBergmeir was created for that specific challenge.
But until now, MSTL was only available in #RStats
#reinventforecasting
Now you can import MSTL from @nixtlainc's StatsForecast open-source library:
Features
π¬ @scikit_learn syntax
β‘οΈ Native support for #spark, @raydistributed, and
@dask_dev
π Predict trend with: ARIMA, Theta, or ETS
#reinventforecasting
Features
π¬ @scikit_learn syntax
β‘οΈ Native support for #spark, @raydistributed, and
@dask_dev
π Predict trend with: ARIMA, Theta, or ETS
#reinventforecasting
MSTL use cases:
π§βπ¬ Run a comparison with popular libraries: github.com/Nixtla/statsfoβ¦
π Forecast next-day electricity demand: nixtla.github.io/statsforecast/β¦
π Predict peaks in the market and reduce electricity bills: nixtla.github.io/statsforecast/β¦
#reinventforecasting
π§βπ¬ Run a comparison with popular libraries: github.com/Nixtla/statsfoβ¦
π Forecast next-day electricity demand: nixtla.github.io/statsforecast/β¦
π Predict peaks in the market and reduce electricity bills: nixtla.github.io/statsforecast/β¦
#reinventforecasting
Stay tuned for tomorrowβs release. π
If you have been waiting for a reproducible comparison between deep learning and statics for time series, you will love it. π₯
#reinventforecasting
If you have been waiting for a reproducible comparison between deep learning and statics for time series, you will love it. π₯
#reinventforecasting
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
γ
