sample at some time vs sample on some interval (~ moving_average)

[sdementen]

I would like to prepare a PR to improve the efficiency of moving_average with pandas=True. I have already the code but I am wondering how to call the function that does this.

More generally, when converting the traces to some regularly spaced pandas time-series, we can either:

• on each specific datetime, give the value of the TimeSeries ==> the current TimeSeries.sample function
• on each interval [datetime, datetime+sampling_period), give the average (or the max, min, median, ...) of the TimeSeries in a similar way as the groupyby().agg() functionality of pandas ==> this can't be called moving_average as a) it does not allow moving averages, b) it is not limited to average (as we have max, min, median, etc).

What name could be use for this "sample on an interval through an 'aggregation' function" ? Would sample_interval makes sense ? would the sample be renamed to sample_instant ?

[stringertheory]

Having a pandas keyword argument on moving_average is another good idea! Thanks for making these issues and pull requests.

On the more general case, I think this might be accomplished by making a function that can do arbitrary convolutions (of which moving average is a special case). That was something that @vlsd mentioned at some point, but we decided against it for the time being since it started to seem too "heavy" for traces in it's current state.

If you're interesting in taking a crack, I think this article is a good resource.

[sdementen]

The need is much more basic than convolutions (which are relatively expensive operations AFAIK). It is just to transform (without moving averages) the TimeSeries to the equivalent pandas Series in a very efficient way. I'll give a shot on a PR on a sample_interval function.