user supplied jacobian

[bjarthur]

this speeds up my particular use case by 3x.

not sure of the best way to expose the interface to curve_fit. for the time being i just left it unchanged.

[bjarthur]

tests are only failing b/c of the Rmath download is stalling.

[pkofod]

Wonderful; finite differences can be really horrible sometimes. This will not only be a speed improvement, but also accuracy improvement in many cases.

edit: btw, please show the functionality in the README.md.

[bjarthur]

another alternative to finite differences is JuliaMath/Calculus.jl#94. i can work that up into a PR if you like. in my use case it was slower, but as you point out, likely more accurate.

any thoughts on hoisting this to the level of curve_fit? the case without weights is easy. not sure what to do with them.

[bjarthur]

updated README, and hoisted interface to curve_fit.

am i right in thinking that the Jacobian simply scales with the weights just like the model does?

[marcusps]

@bjarthur Re: handling weights with the Jacobian, have a look at #45 . If that gets merged first, it should be easy to modify your code to handle the cases with vector and matrix weights.

[blakejohnson]

@bjarthur I merged #45. Could you please fix up the merge conflicts here?

[bjarthur]

@marcusps could you please look this over re. the weights?

@blakejohnson conflicts fixed and tests have passed

[marcusps]

Sure thing — I'll take care of it over the weekend.

…

On Thu, May 4, 2017, 8:13 AM Ben Arthur ***@***.***> wrote: @marcusps <https://github.com/marcusps> could you please look this over re. the weights? @blakejohnson <https://github.com/blakejohnson> conflicts fixed and tests have passed — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#46 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAOKcfmOMnA1Yf5m5cQ7H4cU4DooTyL2ks5r2eq_gaJpZM4MYK1z> .

[pkofod]

this speeds up my particular use case by 3x.

I see you added the new syntax to the README. Could you perhaps also add an example (the test problem for example) to show the performance increase? That would be great, as it makes it much easier for users to see exactly what they're supposed to do.

[pkofod]

Excellent! I have no further comments.

[bjarthur]

example added to README. i also took the liberty to fix a couple deprecation warnings, and to reformat the leading white space in one file.

[bjarthur]

@marcusps is my handling of the weights here kosher? would be nice to merge this PR...

[marcusps]

Looks good to me -- apologies for the delay.