[BUG] ExponentialSmoothing() seems to fit the same model for all time series when ts_num > 1

[morkapronczay]

Describe the bug When using cuml.ExponentialSmoothing results change when modeling varying number of time series ts_num. For the same time series, forecasts differ if more time series are included in the model.

Steps/Code to reproduce bug Follow your own arima demo, but instead of ARIMA, try fitting ExponentialSmoothing() with the same parameters for 2, and 4 variables. I attach a picture and also include the modified notebook in pdf. arima_demo.pdf

Expected behavior This should not have any effect on the fits and the forecast. It seems to me that the same model is fitted on all the time series. For the different time series, separate models should be fitted.

Environment details (please complete the following information):

• Environment location: Docker
• Linux Distro/Architecture: Ubuntu 20.04 amd64
• GPU Model/Driver: RTX 3080 and driver 460.73.01
• CUDA: 11.2
• Method of cuDF & cuML install: Docker:
  • docker pull rapidsai/rapidsai:21.06-cuda11.0-runtime-ubuntu20.04-py3.7 docker run --gpus all --rm -it -p 8888:8888 -p 8787:8787 -p 8786:8786 \ rapidsai/rapidsai:21.06-cuda11.0-runtime-ubuntu20.04-py3.7

Additional context This makes this part quite hard to use, as modeling time series 1-by-1 removes any advantages GPU usage create. It seems to be slower than CPU based pure statsmodels. This is my first Issue on Github, any feedback on it is appreciated!

[vidosits]

Here's a minimal example to reproduce this bug:

import cudf
import numpy as np
from cuml import ExponentialSmoothing
from matplotlib import pyplot as plt

data = cudf.Series([1, 2, 3, 4, 5, 6,
                    7, 8, 9, 10, 11, 12,
                    2, 3, 4, 5, 6, 7,
                    8, 9, 10, 11, 12, 13,
                    3, 4, 5, 6, 7, 8, 9,
                    10, 11, 12, 13, 14],
                    dtype=np.float64)
cu_hw = ExponentialSmoothing(data, seasonal_periods=12, ts_num=1)
cu_hw.fit()
cu_pred = cu_hw.forecast(10)

plt.plot(range(0, 36), data.to_pandas().tolist())
plt.plot(range(36, 46), cu_pred.to_pandas().tolist())

two_series = cudf.concat([data, data*5], axis=1)
cu_hw = ExponentialSmoothing(two_series, seasonal_periods=12, ts_num=2)
cu_hw.fit()
cu_pred = cu_hw.forecast(10)

plt.plot(range(0, 36), two_series.to_pandas()[0].tolist())
plt.plot(range(36, 46), cu_pred.to_pandas()[0].tolist())

minimal_example.ipynb.gz

[dantegd]

cc @Nyrio

[vidosits]

Hey @dantegd , @Nyrio could you take a look at this again if you can? Thanks, in advance!

[github-actions[bot]]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[github-actions[bot]]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[vidosits]

confirmed, still an issue

[github-actions[bot]]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.