Fix minimal installation on globalmodel import

[adamantike]

Since the introduction of globalmodel models, the Kats installation using MINIMAL_KATS=1 fails with the following error:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/src/kats/__init__.py", line 7, in <module>
    from . import detectors  # noqa # usort: skip
  File "/src/kats/detectors/__init__.py", line 20, in <module>
    from .meta_learning import metalearning_detection_model  # noqa # usort: skip
  File "/src/kats/detectors/meta_learning/metalearning_detection_model.py", line 11, in <module>
    from kats.models.metalearner.metalearner_modelselect import MetaLearnModelSelect
  File "/src/kats/models/__init__.py", line 18, in <module>
    from . import globalmodel  # noqa
  File "/src/kats/models/globalmodel/__init__.py", line 5, in <module>
    from . import backtester  # noqa
  File "/src/kats/models/globalmodel/backtester.py", line 16, in <module>
    import torch
ModuleNotFoundError: No module named 'torch'

As the globalmodel/backtester.py file requires torch, it's needed to just warn about the absense of torch, and allow the imports to work for minimal installation.

This behavior was tested using the following imports:

python -c 'from kats.consts import TimeSeriesData'
python -c 'from kats.models.prophet import ProphetModel'
python -c 'from kats.detectors.cusum_detection import CUSUMDetector'
python -c 'from kats.tsfeatures.tsfeatures import TsFeatures'

[michaelbrundage]

Thanks, @adamantike , sorry, just now seeing this PR. I think we've already fixed it -- and added a Github action to test minimal install, so it shouldn't invisibly break again. Is the current code working for you?

[michaelbrundage]

Fixed in c501da1d49313c12616ff06297b221d80cd36f57