Cythonization about pyflux HOT 8 OPEN

https://github.com/jakevdp/cython_template might be useful

from pyflux.

For this task the main bottleneck to optimize are the recursions within _model(). For each model:

ARIMA - the ma recursion
GARCH - the entire recursion
GAS - the entire recursion
GAS Regression - the entire recursion
GP-NARX - no recursions here
Gaussian state space - the kalman filter/smoother recursions
Non-Gaussian state space - should be fine
VAR - the covariance matrix construction

BBVI - the outer for loop
MH - the outer for loop

from pyflux.

That's really useful - seems like the right way to go to ensure Cython isn't mandatory for users. Can't wait to get stuck into this once I've got the other stuff out of the way.

from pyflux.

Tried that template out tonight - works a treat. Will have first Cythonized version this weekend!

from pyflux.

Cool. Glad it worked.

On 19 Jul 2016 9:55 PM, "Ross Taylor" [email protected] wrote:

Tried that template out tonight - works a treat. Will have first
Cythonized version this weekend!

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#5 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AA8DiLVY2xLSmn8GbKTcDc9tNZt9p_PFks5qXTmygaJpZM4I3I-b
.

from pyflux.

Update after recent release:

• GARCH models still need to be Cythonized.
• Need to find good way of getting performance up for GAS/GARCH without compromising the flexibility of allowing users to specify their own assumptions/distributions

from pyflux.

Update: GAS and GASReg models to undergo further Cythonization this weekend.

from pyflux.

GAS models Cythonized further - will be in next release; speedups of 1.5-2x.

from pyflux.