How to call predict function on BART?

Questions
1

I’m trying to figure out how to use the predict function with BART. I know this is very new, and the only example notebook I can find does not use it or do prediction on new data.

I started with the demo notebook and tried adding this cell after completing the trace:

with Model() as model:
BART.predict(BART, X_new=x_test)

But get the error: “AttributeError: type object ‘BART’ has no attribute ‘all_trees’”

Does anyone know how to correctly call this function for predicting on new X data?

Thanks!

1 Like
2

Hi @justindlwhite, there is a bug in the code, to make the predict method works you need to sample a single chain.

1 Like
3

Okay, I tried rerunning the code with trace = pm.sample(cores = 2, chains=1… but am still coming up with same error. Would it be possible to add an example of the predict method to the example notebook?

4

This should work.

with pm.Model() as model:
    μ = pm.BART('μ', X, Y)
    σ = pm.HalfNormal('σ', 1)

    y = pm.Normal('y', μ, σ, observed=Y)
    trace_u = pm.sample(2000, chains=1)
    
μ.distribution.predict(X_new)

Improving the documentation, the API and the models BART is able to run is on my ToDo list.

1 Like
5

Thanks so much, that worked!

6

One clarifying question about this: when I call the .distribution.predict(X_new), is that returning the Y values or the tree distribution μ?

7

It will return the μ. That will change in the future to return Y.

8

Got it, so given your model above, how could I get an updated y value on the X_new?

9

for the moment you have to do it by hand, for example you can approximate it with something like

stats.norm(μ.distribution.predict(X_new), trace_u["σ"].mean())
1 Like
10

Hey, I know this thread is quite old, but I was wondering if you had an idea how to call predict on BART in the newer versions of pymc? I am getting these errors and I am not sure how to sample from the decision tree with new X values now. If you have any suggestions that would be really appreciated!

type(bart)

aesara.tensor.var.TensorVariable

bart.distribution

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Input In [148], in <cell line: 1>()
----> 1 bart.distribution

AttributeError: 'TensorVariable' object has no attribute 'distribution'
1 Like