Retrieve one sample of a probabilistic series as a TimeSeries

[dwolffram]

Hi there,

is there a simple way to retrieve one sample of a probabilistic series as a TimeSeries object? I only found univariate_values() which returns a 1-D Numpy array.

Thanks!

[dennisbader]

Hi @dwolffram, yes there are a couple of ways:

• TimeSeries.random_component_values() returns a 2D array (T, C) with times and random a sample from each component.
• TimeSeries.values(copy=True, sample=0) returns a 2D array (T, C) with times and a sample at index sample from each component.
• TimeSeries.all_values() returns a 3D array (T, C, S) with times, components, and all samples for each time and component.

[dwolffram]

Thanks @dennisbader but these all return a numpy.ndarray and not a TimeSeries.

In the case of a multivariate series, you can do something like series['component_name'] to retrieve a univariate TimeSeries. But there is no way of indexing a stochastic series to retrieve one sample as a TimeSeries, right?

[dennisbader]

Ah sorry, you are right.

You would have to create a new series from the values. with series.with_values(), you can return a new series that has the same components and times as series, but with new values.

E.g.

new_series = series.with_values(series.values(sample=0))

[dwolffram]

Thanks, that was helpful!

However, I got the error ValueError: different number of dimensions on data and dims: 2 vs 3. After adding another dimension it works though:

new_series = series.with_values(np.expand_dims(series.values(sample=0), 1))

[dennisbader]

True 👍