I am dealing with getting different results locally and in databricks most probably from fitting of statsmodels.tsa.holtwinters.ExponentialSmoothing . Based on the documentation, one can't pass directly a random state, but this doesn't guarantee that some implicit random_state doesn't happen down the road, potentially causing the discrepancy. Can you help investigating? Any suggestions are welcome. My ideas include trying np.random.seed(42) random.seed(42) os.environ['PYTHONHASHSEED'] = '42' but it looks a bit ugly. Continue reading...
