Le 3/3/16 06:25, Hernán Morales Durand a écrit :
Hi Stef,
For bioinformatics research it would be **really** nice to have:
- Tree creation algorithms: UPGMA, WPGMA. - MCMC methods: Metropolis-adjusted Langevin algorithm (MALA and SMMALA), Metropolis-coupled MCMC (MC3) would enable Bayesian inferences and to work with evolution of genes. I cannot tell how important is to have such algorithms implemented in Smalltalk. - Mahalanobis distance for analyzing bacterial plasmids. - Random Forest is used a lot for gene expression, protein-protein interaction predictions, genome-wide association studies, etc . For the argument "let's use FFI for it" please remember many of these methods are extended, like enriched random forest, for research with specialized data.
Yes we need random forest for our untangler too.
There are implementations of naive Bayes classifier in NaturalSmalltalk (which could be adapted to SciPharo) http://smalltalkhub.com/#!/~hernan/NaturalSmalltalk <http://smalltalkhub.com/#%21/%7Ehernan/NaturalSmalltalk>
Thanks! First we should clean and move forward SciPharo
As a side note I found an interesting history report of SciPy. From the slide 9 onwards:
http://sssslide.com/www.slideshare.net/shoheihido/sci-pyhistory
Yes this is really interesting. I was wondering what is the NumPy Array exactly page 34 Do you think that this is something like WordArray a kind of Array with typed elements?
Presentation shows SciPy history (credited contributors with pictures, very nice!). As you can se they even have a Foundation NumFOCUS for raising money.
Hernán
2016-03-02 10:17 GMT-03:00 stepharo <stepharo@free.fr <mailto:stepharo@free.fr>>:
Hi guys
I met Didier Besset and we had a great hacking session and discussions with Serge Stinckwich. Didier would like to help Pharo and the numerical part of it. ***Big thanks*** Didier.
We would like to do several things:
- Work on "Hows to" The numerical methods in Pharo is good but the gap between us and the math is too large :) so the idea is to have a series of "how to ..." - histomgram (simple, based on distribution)
- Improve the SciPharo/NumPha (previously SciSmalltalk) library This morning we started to implement a ComponentPrincipalDecomposition by combining two of the objects available in SciPharo. Then we started to enhance the distributions to make sure that we can plug other distribution for having controlled random number.
- Do a public call to know what is missing for you: this is this mail :) Didier would like to work on concrete cases. I love that attitude So tell us :)
Hernan ?? Alex: ?? pvalue? better distribution? Vincent: covariance? CPA? Philippe: times series Serge R frames? Sami: Better random number and various distributions?
- Organise a two day lectures with practices on concrete case in September with a recording session. Either at IRD Bondy or Lille.
- Make sure that the Numerical Method book will get on lulu :) with a better cover and title :)
Stef