astro-informatics / harmonic Goto Github PK

add_chains_2d_list(add chain in 2d array with list specifying the starts and ends)
and get_sub_chains (makes to chain instance pulling out certain chains)
functionality added to chain class to allow easy use in data handling module

Apparently we have been using a trial period of Travis CI, which is about to come to an end.

Need to set up to use educational github account so can use Travis CI for free with a private repo.

Although @cgrwallis has already implemented this it still needs to be reviewed.

Should store samples from multiple chains in single 2D numpy.ndarray with 1D chain starts index vector (with Nchains+1) elements where last element is total number of samples.

The user should specify the final number of blocks (i.e. new chains) and pass a Chains object with data initialised (can contain one or multiple chains).

We split each existing chain into new blocks, proportionally to the size of the current chains. Final blocks will end up contain slightly different numbers of samples (we don't ever want to throw away samples).

The main thing is that the blocks are independent.

Seems like most of our errors should be ValueError (see https://stackoverflow.com/questions/28514501/is-valueerror-an-appropriate-exception-to-raise-if-user-passes-arguments-of-diff).

More info on possible exceptions here: https://docs.python.org/3/library/exceptions.html

Add abstract method is_fitted to model.

In evidence module, check model is fitted.

At present the centre is set by the mean of the samples but this could instead be fitting during hyperparameter optimisation.