SamplesDict class

pesummary handles the posterior samples through a custom SamplesDict class. This SamplesDict class is inherited from the core dict class and includes extra properties for returning things like the maximum likelihood samples, median for each marginalized posterior, number of samples stored etc. The marginalized posterior is stored in a pesummary.utils.utils.Array object. For details about this class see Array class

Initializing the SamplesDict class

The SamplesDict class is initialized with an array of parameters and an array of samples for each parameter,

>>> from pesummary.utils.samples_dict import SamplesDict
>>> parameters = ["a", "b"]
>>> samples = [[1,2,3,4,5,6], [6,5,4,3,2,1]]
>>> samplesdict = SamplesDict(parameters, samples)
>>> print(samplesdict)
idx     a              b
0       1.000000       6.000000
1       2.000000       5.000000
2       3.000000       4.000000
3       4.000000       3.000000
4       5.000000       2.000000
5       6.000000       1.000000
>>> samplesdict = SamplesDict({"a": [1,2,3,4,5,6], "b": [6,5,4,3,2,1]})
>>> print(samplesdict)
idx     a              b
0       1.000000       6.000000
1       2.000000       5.000000
2       3.000000       4.000000
3       4.000000       3.000000
4       5.000000       2.000000
5       6.000000       1.000000

Alternatively, it may be initialized with the path to a result file containing posterior samples,

>>> samplesdict = SamplesDict.from_file("path_to_file.hdf5")

This classmethod simply calls the read function and initializes the class with the parameters and samples that are already stored.

Using the SamplesDict properties

Below we show some of the useful properties of the SamplesDict class. For full details see the doc string,

>>> samplesdict.minimum
{'a': Array([1]), 'b': Array([1])}
>>> samplesdict.maximum
{'a': Array([6]), 'b': Array([6])}
>>> samplesdict.median
{'a': Array([3.5]), 'b': Array([3.5])}
>>> samplesdict.mean
{'a': Array([3.5]), 'b': Array([3.5])}
>>> samplesdict.number_of_samples
6

Using the SamplesDict functions

The SamplesDict class also has functions to try and make it easier to manipulate the stored posterior samples. These functions include conversions to other data structures, downsampling and even removing a set of posterior samples,

>>> numpy_structured_array = samplesdict.to_structured_array()
>>> numpy_structured_array
rec.array([(1., 6.), (2., 5.), (3., 4.), (4., 3.), (5., 2.), (6., 1.)],
      dtype=[('a', '<f8'), ('b', '<f8')])
>>> sliced = samplesdict[:2]
>>> sliced
{'a': Array([1, 2]), 'b': Array([6, 5.])}
>>> pandas_dataframe = samplesdict.to_pandas()
>>> pandas_dataframe

>>> downsample = samplesdict.downsample(2)
>>> downsample
{'a': Array([1, 6]), 'b': Array([6, 1])}
>>> discard = samplesdict.discard_samples(3)
>>> discard
{'a': Array([4, 5, 6]), 'b': Array([3, 2, 1])}

The SamplesDict class also provides the ability to plot the posterior samples directly. This can be achieved through the plot() method. For example, if we want to plot the samples as a KDE, we can run,

>>> parameter = "a"
>>> fig = samplesdict.plot(parameter, type="hist", kde=True)
>>> fig.show()

To see the full list of available plots, you can run:

>>> samplesdict.available_plots

To see the list of args and kwargs for each plot type, you can run:

>>> help(samplesdict.plot)