Last active
April 13, 2021 20:27
-
-
Save twiecki/a77104299535b64b58953de3c84df56f to your computer and use it in GitHub Desktop.
stochastic_volatility.ipynb
This looks like something we need to update in PyMC3, as well.
Here's a quick comparison of the timing with and without graph optimizations (the example/model is taken from this notebook):
fgraph = model.logp.f.maker.fgraph
...
get_ipython().run_line_magic('timeit', 'fn2(*init_state).block_until_ready()')
# 198 µs ± 18.3 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
fgraph = aesara.graph.fg.FunctionGraph(model.free_RVs, [model.logpt])
...
get_ipython().run_line_magic('timeit', 'fn2(*init_state).block_until_ready()')
# 236 µs ± 10.7 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I suppose
pm.sample()already does this?