Change seasonality of "D" to 7, and 52 for "W"

[AaronSpieler]

Description

• currently we have DEFAULT_SEASONALITIES = {"H": 24, "D": 1, "W": 1, "M": 12, "B": 5}
• however, having a seasonality of 1 has no real benefit to my knowledge (please correct me if I'm mistaken)
  • I base my knowledge on NBEATSEstimator and Naive2Predictor
• therefore I suggest seasonality of 7 for Daily data "D"
  • to me this makes sense since this would capture the seasonality of workdays and weekends and in general the work week
• additionally we could think of a seasonality of 52 (or 4) for weekly data "W"

Examples

• Wavenet:

              {
                  "H": 7 * 24,
                  "D": 7,
                  "W": 52,
                  "M": 12,
                  "B": 7 * 5,
                  "min": 24 * 60,
              },

• DeepState:

  FREQ_LONGEST_PERIOD_DICT = {
  "M": 12,  # yearly seasonality
  "W-SUN": 52,  # yearly seasonality
  "D": 31,  # monthly seasonality
  "B": 22,  # monthly seasonality
  "H": 168,  # weekly seasonality
  "T": 1440,  # daily seasonality
  }

[kaijennissen]

Any plans to change this in the near future? Because I recently worked with the SeasonalNaivePredictor and daily observations and was surprised that I had to manually set season_length=7 despite the call to get_seasonality(freq) in the __init__-method.

[jaheba]

I agree that we should have a daily seasonality of 7. Are there cases that we are missing, i.e. why was 1 chosen in the first place?

However, my feeling is that there is not such a fitting seasonality for weekly-frequency.