-
Notifications
You must be signed in to change notification settings - Fork 1
Past sprints
We are organizing a sprint before the PyconFR 2012 conference.
- Nelle Varoquaux - Isotonic regression
- Olivier Grisel (working with Gaël Varoquaux on joblib parallelism)
- Alexandre Gramfort
- Fabian Pedregosa (Things I could work on: Implement ranking algorithms (RankSVM, IntervalRank), help with the isotonic regression and group lasso pull request)
- Bertrand Thirion
- Gaël Varoquaux (working with Olivier Grisel on joblib parallelism)
- Alexandre Abraham
- Virgile Fritsch
- Nicolas Le Roux (providing machine learning expertise for RBM and DBN coding)
La Villette, Paris, the 13th & 14th of September, from 10:00 until 18:00. The sprint will take place in the 'Carrefour Numerique', floor -1 of the 'cité des sciences': http://www.pycon.fr/2012/venue/
Top priorities are merging: pull requests, fixing easyfix issues and improving documentation consistency.
In addition to the tasks listed below, it is useful to consider any issue in this list : https://github.com/scikit-learn/scikit-learn/issues
- Improve test coverage: Run 'make test-coverage' after installing the coverage module, find low hanging fruits to improve coverage, and add tests. Try to test the logic, and not simple aim for augmenting the number of lines covered.
- Finish estimator summary PR: https://github.com/scikit-learn/scikit-learn/pull/804
Improving and merging existing pull requests is the number one priority: https://github.com/scikit-learn/scikit-learn/pulls
There is a lot of very good code lying there, it often just needs a small amount of polishing
- Affinity propagation using sparse matrices: the affinity propagation algorithm (scikits.learn.cluster.affinity_propagation_) should be able to work on sparse input affinity matrices without converting them to dense. A good implementation should make this efficient on very large data.
- Improve the documentation: You understand some aspects machine-learning. You can help making the scikit rock without writing a line of code: http://scikit-learn.org/dev/developers/index.html#documentation. See also Documentation-related issues in the issue tracker.
- Text feature extraction (refactoring / API simplification) + hashing vectorizer: Olivier Grisel
- Nearest Neighbors Classification/Regression : allowing more flexible Bayesian priors (currently only a flat prior is used); implementing alternative distance metrics: Jake Vanderplas
- Group Lasso: Continue with pull request https://github.com/scikit-learn/scikit-learn/pull/947. Participants: @fabianp
Participants: @mblondel
- Code clean up
- Speed improvements: don't reallocate clusters, track clusters that didn't change, triangular inequality
- L1 distance: use L1 distance in e step and median (instead of mean) in m step
- Fuzzy K-means: k-means with fuzzy cluster membership (not the same as GMM)
- Move argmin and average operators to pairwise module (for L1/L2)
- Support chunk size argument in argmin operator
- Merge @ogrisel's branch
- Add a score function (opposite of the kmeans objective)
- Sparse matrices
- fit_transform
- more output options in transform (hard, soft, dense)
Participants: @larsmans
- EM algorithm for Naive Bayes (there is a pull request linguering)
- Fix utility code to handle partially labeled data sets
-
- Patch liblinear to have warm restart + LogisticRegressionCV.
- Comment (by Fabian): I tried this, take a look here: liblinear fork
- Locality Sensitive Hashing, talk to Brian Holt
- Fused Lasso
- Group Lasso, talk to Alex Gramfort (by email), or Fabian Pedregosa
- Manifold learning: improve MDS (talk to Nelle Varoquaux), t-SNE (talk to DWF)
- Sparse matrix support in dictionary learning module
- Jaques Grobler
- Fabian Pedregosa. I'll be working on improving test coverage and implementing Group Lasso. Also, I can introduce newcomers into the scikit-learn workflow.
- Université libre de Bruxelles
Contributors might find useful the coding guidelines .
We are organizing a coding sprint after the NIPS 2011 conference.
For this sprint, we are trying to gather funding for contributors to fly in. Please list your name and who is funding your trip.
- Gael Varoquaux: Funding: INRIA
- Bertrand Thirion: Funding: INRIA
- Fabian Pedregosa: Funding: INRIA
- Alex Gramfort: Funding: INRIA
- Olivier Grisel: Funding: Google + tinyclues
- Jake Vanderplas: Funding: Google + tinyclues
- David Warde-Farley: Funding: LISA
- Gilles Louppe: Funding: University of Liège
- Lars Buitinck: Funding: Google + tinyclues
- Vlad Niculae: Funding: Google + tinyclues
- Andreas Mueller: Funding: Google + tinyclues
- Mathieu Blondel: Funding: Google + tinyclues + private
- Nicolás Della Penna: private.
- Granada University, Instituto de la Paz y los Conflictos, Centro de Documentación Científica de la Universidad de Granada, first floor Campoamor classroom (map) , from 10:00 until 18:00
Contributors might find useful the coding guidelines .
Top priorities are merging: pull requests, fixing easyfix issues and improving documentation consistency.
In addition to the tasks listed below, it is useful to consider any issue in this list : https://github.com/scikit-learn/scikit-learn/issues
- Merge in Randomized linear models (branch 'randomized_lasso' on GaelVaroquaux's github (Gael Varoquaux and Alex Gramfort working on this)
- Improve test coverage: Run 'make test-coverage' after installing the coverage module, find low hanging fruits to improve coverage, and add tests. Try to test the logic, and not simple aim for augmenting the number of lines covered.
- Py3k support: First test joblib on Python3, then scikit-learn. Both generate sources that are python3 compatible, but these have not been tested.
Improving and merging existing pull requests is the number one priority: https://github.com/scikit-learn/scikit-learn/pulls
There is a lot of very good code lying there, it often just needs a small amount of polishing
- Rationalize images in documentation: we have 56Mo of images generated in the documentation (doc/_build/html/_images). First we should save jpg instead of pngs: it shrinks this directory to 45Mo (not a huge gain, granted). Second there is many times the same file saved. I need to understand what is going on, and fix that.
- Affinity propagation using sparse matrices: the affinity propagation algorithm (scikits.learn.cluster.affinity_propagation_) should be able to work on sparse input affinity matrices without converting them to dense. A good implementation should make this efficient on very large data.
- Improve the documentation: You understand some aspects machine-learning. You can help making the scikit rock without writing a line of code: http://scikit-learn.org/dev/developers/index.html#documentation. See also Documentation-related issues in the issue tracker.
- Text feature extraction (refactoring / API simplification) + hashing vectorizer: Olivier Grisel
- Nearest Neighbors Classification/Regression : allowing more flexible Bayesian priors (currently only a flat prior is used); implementing alternative distance metrics: Jake Vanderplas
Participants: @mblondel
- Code clean up
- Speed improvements: don't reallocate clusters, track clusters that didn't change, triangular inequality
- L1 distance: use L1 distance in e step and median (instead of mean) in m step
- Fuzzy K-means: k-means with fuzzy cluster membership (not the same as GMM)
- Move argmin and average operators to pairwise module (for L1/L2)
- Support chunk size argument in argmin operator
- Merge @ogrisel's branch
- Add a score function (opposite of the kmeans objective)
- Sparse matrices
- fit_transform
- more output options in transform (hard, soft, dense)
Participants: @mblondel
- Merge random SVD PR
- Merge sparse RP PR
- Cython utils for fast and memory-efficient projection
Participants: @amueller
- Move to random projection module
Participants: @vene
- Fix (document) alpha scaling
- Merge SparseCoder pull request
- Merge KMeansCoder pull request
- Begin work on supervised image classification
Participants: @larsmans
- EM algorithm for Naive Bayes
- Fix utility code to handle partially labeled data sets
-
- Patch liblinear to have warm restart + LogisticRegressionCV.
- Comment (by Fabian): I tried this, take a look here: liblinear fork
- Decision Tree (support boosted trees, loss matrix, multivariate regression)
-
- Ensemble classifiers
- Comment (by Gilles): I plan to review @pprett PR on Gradient Boosted Trees. I also want to implement parallel tree construction and prediction in the current implementation of forest of trees.
- Locality Sensitive Hashing, talk to Brian Holt
- Fused Lasso
- Group Lasso, talk to Alex Gramfort (by email)
- Manifold learning: MDS, t-SNE (talk to DWF)
- Bayesian classification (e.g. RVM)
- Sparse matrix support in dictionary learning module
Some of us are planning to stay at a Guest House in Granada to reduce the Hotel costs. If you are interested add your name and arrival and departure dates below:
| Name | From | To |
|---|---|---|
| Olivier Grisel | Dec. 11 | Dec. 21 |
| Gael Varoquaux | Dec. 11 | Dec. 21 |
| David Warde-Farley | Dec. 18 | Dec. 21 |
| Alex Gramfort | Dec. 11 | Dec. 21 |
| Jake Vanderplas | Dec. 15 | Dec. 22 |
| Bertrand Thirion | Dec 12 | Dec. 20 |
| Gilles Louppe | Dec 18 | Dec. 21 |
| Mathieu Blondel | Dec 18 | Dec. 22 |
| Lars Buitinck | Dec 18 | Dec 22 |
| Vlad Niculae | Dec 18 | Dec 22 |
| Andreas Mueller | Dec 11 | Dec 22 |
| Nicolás Della Penna | Dec 18 | Dec 22 |
| (add your name here) |
We are organizing a coding sprint the days before EuroScipy 2011
- Olivier Grisel: review code (esp. related to Vlad's GSoC), doc improvements, maybe work on finalizing Power Iteration Clustering or the text feature extraction
- Gael Varoquaux: merging pull requests
- Vlad Niculae: merge remaining DictionaryLearning code, doc improvements, maybe work on SGD matrix fact. w/ someone?
- Satra Ghosh: work on the ensemble/tree/random forest (only on the 24th)
- Brian Holt: tree and random forest code, improve test coverage, doc improvements
- Bertrand Thirion: reviewing GMM and related stuff or manifold learning (probably 24th only).
- Ralf Gommers: work on joblib (only 24th, from ~12.00)
- Vincent Michel: work on bi-clustering, doc improvements, code review.
- Mathieu Blondel: multi-class reductions (only 24th, GMT+9)
- Fabian Pedregosa : strong-rules for coordinate descent, grouped lasso or related stuff, py3k support.
- Alexandre Gramfort : reviewing commits and sending negative comments to harass Fabian while he is away because he kind of likes that
- Jean Kossaifi
- Virgile Fritsch (only 24th): working on issues (pairwise distances, incompatibility with scipy 0.8, ...) and pull requests merging.
- In Paris: at ENS, in the physics department (24 rue Lhomond), probably in some classrooms on the 3rd floor.
Location At the scipy conference (Austin)
- Gael Varoquaux: review code, merge
- Marcel Caraciolo: review code, easyfix issues.
- David Warde-Farley: review
- In Paris: at Logilab's (104 boulevard blanqui, Paris) - Metro 6 - Glacière
- In Boston at MIT (36-537: 5th floor of building 36)
- On IRC (#scikit-learn on irc.freenode.net)
Please add skills/interests or planned task, to facilitate the sprint organization and pairing of people on tasks. To share knowledge as much as possible, it would be ideal to have pair-like programming of 2 people on a task, with different skills.
At Logilab, Paris (from 9H to 19H):
- Gaël Varoquaux: task: code review, pair programming on specific task where needed.
- Julien Miotte
- Feth Arezki: could help with coding (w/ the logger?), LaTeX. Interested in learning about scikit.
- Nelle Varoquaux: task: minibatch k-means
- Fabian Pedregosa
- Vincent Michel: task: code review, pair programming. features: ward's clustering.
- Luis Belmar-Letelier
- Thouis Jones: task: BallTree cython wrapper, documentation, whatever.
At MIT, Boston:
- Alexandre Gramfort: task: code review and pair programming
- Demian Wassermann: task: Gaussian Processes with sparse data
- Satra Ghosh: task: Ensemble Learning, random forests
- Nico Pinto
- Pietro Berkes
At IRC (from around 9am Brasília time (GMT-3):
- Alexandre Passos: task: dirichlet process mixture of gaussian models (In progress)
- Vlad Niculae: task: matrix factorization (In progress)
- Marcel Caraciolo: task: help in docs and bug fixes (beginner in the project).
Place:
INRIA research center in Saclay-Ile de France, also in channel #scikit-learn, on irc.freenode.org. Room to be determined.
Some ideas:
- extend the tutorial with features selection, cross-validation, etc
- design a sphinx template for the main web page [here http://www.flickr.com/photos/fseoane/4573612893/] is a temptative design, but was not translated into a sphinx template.
- Group lasso with coordinate descent in GLM module
- Covariance estimators (Ledoit-Wolf) -> Regularized LDA
- Add transform in LDA
- PCA with fit + transform
- preprocessing routines (center, standardize) with fit transform
- K-means with Pybrain heuristic
- Make Pipeline object work for real
- FastICA
Anything you can think of, such as:
- Spectral Clustering + manifold learning (MDS/PCA, Isomap, Diffusion maps, tSNE)
- Canonical Correlation Analysis
- Kernel PCA
- Gaussian Process regression
Place:
channel #scikit-learn, on irc.freenode.org. If you do not have an IRC client or behind a firewall, check out http://webchat.freenode.net/
Some ideas:
- adapt the plotting features from the em module into gmm module.
- incorporate more datasets : the diabetes from the lars R package, featured datasets from http://archive.ics.uci.edu/ml/datasets.html , etc.
- anything from the issue tracker.
- extend the tutorial with features selection, cross-validation, etc
- profile and improve the performance of the gmm module.
- submit some new classifier
- refactor the ann module (artificial neural networks) to conform to the API in the rest of the modules, or submit a new ann module.
- make it compatible with python3 (shouldn't be hard now that there's a numpy python3 relase)
- design a sphinx template for the main web page [here http://www.flickr.com/photos/fseoane/4573612893/] is a temptative design, but was not translated into a sphinx template.
- anything you can think of.
Place:
channel #learn, on irc.freenode.org. If you do not have an IRC client or behind a firewall, check out http://webchat.freenode.net/
Possible Tasks:
- Document our design choices (methods in each class, convention for estimated parameters, etc.). Most of this is in ApiDiscussion.
- Documentation for neural networks (nonexistent)
- Examples. We currently only have a few of them. Expand and integrate them into the web page.
- Write a Tutorial.
- Write a FAQ.
- Documentation and Examples for Support Vector Machines. What's in the web is totally outdated. Integrate the documentation from gumpy, see ticket:27 (assigned: Fabian Pedregosa)
- Review documentation.
- Customize the sphinx generated html.
- Create some cool images/logos for the web page.
- Create some benchmark plots.
Terminated, see http://fseoane.net/blog/2010/scikitslearn-coding-spring-in-paris/
- Alexandre Gramfort
- Olivier Grisel
- Vincent Michel
- Fabian Pedregosa
- Bertrand Thirion
- Gaël Varoquaux
Goals
Implement a few targeted functionalities for penalized regressions.
Target functionalities
- GLMnet
- Bayesian Regression (Ridge, ARD)
- Univariate feature selection function
Edouard: Most of things we need are already in datamind, the main main issue is to cut the dependance with FFF(nipy)
Extras, if time permits:
- LARS
Proposed workflow
Pair programming:
- GLMNet (AG, OG)
- Bayesian regression (FP, VM)
- Feature selection (BT, GV)
- LARS: Whoever is finished first.
Place in the repository
- I think GLMNet goes well in scikits.learn.glm.
Edouard: The GLM term is confusing: Indeed in GLMNet the G means "generalized", however in neuroimaging people understand "general" which is in fact a linear model
- Bayessian regression: scikits.learn.bayes . It's short and explicit.
Past sprints
Place:
INRIA research center in Saclay-Ile de France, also in channel #scikit-learn, on irc.freenode.org. Room to be determined.
Some ideas:
- extend the tutorial with features selection, cross-validation, etc
- design a sphinx template for the main web page [here http://www.flickr.com/photos/fseoane/4573612893/] is a temptative design, but was not translated into a sphinx template.
- Group lasso with coordinate descent in GLM module
- Covariance estimators (Ledoit-Wolf) -> Regularized LDA
- Add transform in LDA
- PCA with fit + transform
- preprocessing routines (center, standardize) with fit transform
- K-means with Pybrain heuristic
- Make Pipeline object work for real
- FastICA
Anything you can think of, such as:
- Spectral Clustering + manifold learning (MDS/PCA, Isomap, Diffusion maps, tSNE)
- Canonical Correlation Analysis
- Kernel PCA
- Gaussian Process regression
Place:
channel #scikit-learn, on irc.freenode.org. If you do not have an IRC client or behind a firewall, check out http://webchat.freenode.net/
Some ideas:
- adapt the plotting features from the em module into gmm module.
- incorporate more datasets : the diabetes from the lars R package, featured datasets from http://archive.ics.uci.edu/ml/datasets.html , etc.
- anything from the issue tracker.
- extend the tutorial with features selection, cross-validation, etc
- profile and improve the performance of the gmm module.
- submit some new classifier
- refactor the ann module (artificial neural networks) to conform to the API in the rest of the modules, or submit a new ann module.
- make it compatible with python3 (shouldn't be hard now that there's a numpy python3 relase)
- design a sphinx template for the main web page [here http://www.flickr.com/photos/fseoane/4573612893/] is a temptative design, but was not translated into a sphinx template.
- anything you can think of.
Documentation Week, 14-18 March 2010
Place:
channel #learn, on irc.freenode.org. If you do not have an IRC client or behind a firewall, check out http://webchat.freenode.net/
Possible Tasks:
- Document our design choices (methods in each class, convention for estimated parameters, etc.). Most of this is in ApiDiscussion.
- Documentation for neural networks (nonexistent)
- Examples. We currently only have a few of them. Expand and integrate them into the web page.
- Write a Tutorial.
- Write a FAQ.
- Documentation and Examples for Support Vector Machines. What's in the web is totally outdated. Integrate the documentation from gumpy, see ticket:27 (assigned: Fabian Pedregosa)
- Review documentation.
- Customize the sphinx generated html.
- Create some cool images/logos for the web page.
- Create some benchmark plots.
Terminated, see http://fseoane.net/blog/2010/scikitslearn-coding-spring-in-paris/
- Alexandre Gramfort
- Olivier Grisel
- Vincent Michel
- Fabian Pedregosa
- Bertrand Thirion
- Gaël Varoquaux
Implement a few targeted functionalities for penalized regressions.
- GLMnet
- Bayesian Regression (Ridge, ARD)
- Univariate feature selection function
Edouard: Most of things we need are already in datamind, the main main issue is to cut the dependance with FFF(nipy)
Extras, if time permits:
- LARS
Pair programming:
- GLMNet (AG, OG)
- Bayesian regression (FP, VM)
- Feature selection (BT, GV)
- LARS: Whoever is finished first.
- I think GLMNet goes well in scikits.learn.glm.
Edouard: The GLM term is confusing: Indeed in GLMNet the G means "generalized", however in neuroimaging people understand "general" which is in fact a linear model
- Bayessian regression: scikits.learn.bayes . It's short and explicit.
Edouard: Again the term Bayes might not lead to a clear organization of algorithms.
- Feature selection: featsel? selection ? I'm not sure about this one.
AG : maybe univ?
Edouard: Maybe it is to early to decide the structure of the repository during your coding sprint. I think this organization should follow discussion we had we Fabian, Gael and Bertand. Next I tried to synthesize those discussions, however its just a proposition and many things are missing:
If there's code that we want to share and it does not fit into any of these schemes, it's ok to put it into sandbox/ (it does not yet exist)
- Feature selection: featsel? selection ? I'm not sure about this one.
AG : maybe univ?
Edouard: Maybe it is to early to decide the structure of the repository during your coding sprint. I think this organization should follow discussion we had we Fabian, Gael and Bertand. Next I tried to synthesize those discussions, however its just a proposition and many things are missing: