pm.Minibatch Doc string

Questions v5
1

I am (re)-learning to use ADVI with PYMC and was working through the example here : Introduction to Variational Inference with PyMC — PyMC example gallery
In that example, the authors show a rather extensive docstring for pm.Minibatch but the docstring for Minibatch now is just the anemic:

Get random slices from variables from the leading dimension.

    Parameters
    ----------
    variable: TensorVariable
    variables: TensorVariable
    batch_size: int

    Examples
    --------
    >>> data1 = np.random.randn(100, 10)
    >>> data2 = np.random.randn(100, 20)
    >>> mdata1, mdata2 = Minibatch(data1, data2, batch_size=10)

Is the original doc string incorrect? For example is this still correct : “Importantly, we need to make PyMC “aware” that a minibatch is being used in inference. Otherwise, we will get the wrong :math:logp for the model. the density of the model logp that is affected by Minibatch. See more in the examples below. To do so, we need to pass the total_size parameter to the observed node, which correctly scales the density of the model logp that is affected by Minibatch.”
If so is this documented now somewhere else?

2

Yes, you need to pass total_size, you should see a warning when you call pm.fit if you don’t.

I agree the docstring is pretty sad. PRs welcome :slight_smile:

1 Like