Analyze

pyggdrasil.analyze.rhats(chains)

Compute estimate of rank normalized split R-hat for a set of chains.

Sometimes referred to as the potential scale reduction factor / Gelman-Rubin statistic.

Used the “rank” method recommended by Vehtari et al. (2019)

The rank normalized R-hat diagnostic tests for lack of convergence by comparing the variance between multiple chains to the variance within each chain. If convergence has been achieved, the between-chain and within-chain variances should be identical.

Parameters:

Name	Type	Description	Default
chains	ndarray	array of arrays to calculate R-hat for (minimum of 2 chains, minimum of 4 draws)	required

Returns:

Type	Description
ndarray	R-hat for given chains from index 4 to length,
ndarray	returns list that is 4 shorter than the length of the chains

Note

• May return NaN if the chains are too short and all values are the same
• May raise Out of memory error if the chains are too long, 100000 samples still works, 1000000 does not - storing all truncated chains in memory is too much even at 24 GB RAM ?! TODO: find a way to calculate R-hat for longer chains

pyggdrasil.analyze.ess(chains)

Calculates the effective sample size of a set of chains.

used the “bulk” method recommended by Vehtari et al. (2019) rank normalized draws are used to calculate the effective sample size

chains: array of arrays to calculate ESS for (minimum of 2 chains, minimum of 4 draws)

Returns:

Type	Description
tuple[ndarray, ndarray]	ess_bulk, ess_tail for given chains from index 4 to length,

pyggdrasil.analyze.Metrics

Metrics for comparing trees.

Attributes:

Name	Type	Description
_METRICS	Dict[str, Callable[[TreeNode, TreeNode], float]]	Dictionary of metrics.

get(metric) staticmethod

Return metric function.

Parameters:

Name	Type	Description	Default
metric	str	Name of metric.	required

Returns:

Type	Description
Callable[[TreeNode, TreeNode], float]	AD: Ancestor-Descendant Similarity; pyggdrasil.distances.AncestorDescendantSimilarity().calculate,
Callable[[TreeNode, TreeNode], float]	MP3: MP3 Similarity; pyggdrasil.distances.MP3Similarity().calculate,
Callable[[TreeNode, TreeNode], float]	TrueTree: True Tree Similarity; pyggdrasil.compare_trees
Callable[[TreeNode, TreeNode], float]	DL: Different Lineage Similarity; pyggdrasil.distances.DifferentLineageSimilarity().calculate,
Callable[[TreeNode, TreeNode], float]	MLTD: MLTD Similarity; pyggdrasil.distances.MLTDSimilarity().calculate,

pyggdrasil.analyze.to_pure_mcmc_data(mcmc_samples)

Converts McmcRunData to PureMcmcData.

Takes a list of MCMCSamples converts it into a xarray easy to plot.

Parameters:

Name	Type	Description	Default
mcmc_samples		list[MCMCSample] - list of MCMC samples	required

Returns: PureMcmcData

pyggdrasil.analyze.check_run_for_tree(desired_tree, mcmc_samples)

Check if a tree is in an MCMC run.

Returns of list of tuples of (iteration, tree, log-probability), or False. Goes through entire chain to find all instances of the tree.

Parameters:

Name	Type	Description	Default
desired_tree		Tree	required
mcmc_samples		McmcRunData	required

Returns: bool or list(tuple[int, Tree, float])

pyggdrasil.analyze.analyze_mcmc_run(mcmc_data, metric, base_tree)

Analyze a MCMC run.

Parameters:

Name	Type	Description	Default
mcmc_data		PureMcmcData MCMC run data to analyze of iteration no., tree, and log-probability	required
metric		Callable[[TreeNode, TreeNode], float] metric to apply to the trees.	required
base_tree		TreeNode Tree to compare all applicable metrics to.	required

Returns:

Type	Description
list[int]	list[int], list[float]
list[float]	Iteration number and results of the metric.