Pedro Capelastegui

I have had some trouble installing Anticipy on M1 Macs, because of libraries like SciPy. We should add a note about this in the documentation. I found a solution in...

Implement the following improvements for model_spike: - cache variables, see https://github.com/sky-uk/anticipy/issues/197 - new model: model_ramped_spike model_spike - review code - can make more similar to spike_date model_spike_date - review code...

Error: Weekly Naive model used on series < 7 samples

1

On #195 , we introduced auto detection of appropriate naive models. Unfortunately, this breaks for series without a full week worth of samples - a weekly naive model is used...

On daily-sampled series, try running weekly forecasts, then upscaling

2

We have found that, in some cases, for the same time series, a weekly sampled forecast may perform better than a daily sampled one. We could try, for daily sampled...

Support reproducible scenarios with random seed

1

Anticipy forecasts are non-deterministic, which is a problem for some cases. We should test and document how to get reproducible forecasts using numpy.random.seed()

Implement function to check whether a seasonality model should be considered

2

Currently, we consider too many model combinations for series with daily samples: - weekly and yearly seasonality - additive and multiplicative models - multiple trend and seasonality models We take...

``` anticipy.forecast_plot.plot_forecast(df_data.head(0),width=1600, height=900, output='jupyter') ``` ``` ZeroDivisionError: float division by zero ``` We should replace this with a better error message. We could add the following line: ``` assert not...

forecast_plot.plot_forecast() uses faceted plots whenever the input data has multiple source ID's. However, there is a minor bug in the logic to determine this: currently, the `subplots` variable is true...

In forecast_plot, add support for plotting forecast inputs

1

The plotting functions in module forecast_plot generate plots for our forecast outputs. It would be convenient if we were also able to generate plots for our forecast inputs, so that...

Forecast logic sometimes skip samples (e.g. outliers) set to weight=0. Our plots should show this somehow. Ideally, by having different alpha values, if that won't mess with plotly grouping.