Pedro Capelastegui

 When using multiplicative composition, we sometimes get model coefficients that result in values equal or lower than zero. The plot above shows a model `((linear+ramp)*(season_month*calendar_uk))` where negative coefficients in...

We should explore different methods for outlier detection. Library pyod (https://github.com/yzhao062/pyod) looks promising. Things to consider: - Including dummy models as variables for outlier detection. This would allow us to...

In some cases, we may want to use anticipy with input time series that have very few samples. Regression models can fail in these scenarios, so it would be good...

In some time series, we have found trends changing for weekdays but not for weekends, or vice versa. Having different linear trends for each could help with this. 

Samples with weight=0 should be ignored by this function.

If first sample of series is outlier, get_model_outliers won't detect it.

We need an objective way to evaluate our forecast accuracy. To that end, we should generate forecasts for a well established dataset. We will use the M3 and M4 forecasting...

On #100 , we implemented a new model to support event calendars, including a calendar of holidays for the UK and US. However, we are missing a tutorial section showing...

Currently, a new user importing anticipy and running it on a clean environment will not see any logs. In order to enable logging, they need to call: ``` logging.basicConfig(level=logging.INFO) ```...

Currently, the output of a forecast is non-deterministic: running it multiple times will get different results. In some applications, we may need a reproducible process, so that the same inputs...