Machine Learning Cheat Sheet Map
scikit-learn.orgJust a note. scikit has logistic regression, but it isn't shown at the top level. It is contained within `LinearSVC` on the diagram: http://scikit-learn.org/stable/modules/generated/sklearn.lin...
It's currently down - is it the same one discussed at some length a year ago?
https://news.ycombinator.com/item?id=5831512
Here are some other resources for machine learning, not necessarily restricted to the algorithms implemented in SciKit:
http://eferm.com/wp-content/uploads/2011/05/cheat3.pdf
http://peekaboo-vision.blogspot.ca/2013/01/machine-learning-...
http://rise.cse.iitm.ac.in/wiki/index.php/Introduction_to_Ma...
"Mathbabe" is currently working on something similar: http://mathbabe.org/2014/08/28/a-decision-tree-for-decision-... even if she has just started working on this (and she mentions the OP item herself).
Dimensionality reduction can be a goal in and of itself, but many of the same techniques (i.e. feature selection) are useful precursors for classification, clustering, and regression. It would be nice to capture that on the diagram. More arrows, please. :)
The "text data" decision point seems arbitrary and in my opinion, not useful. I've analyzed text data with a Naive Bayes a classifier as well as SVM. I really like what the chart is trying to be, but I think it is editorializing too much.
That entire map is complete and utter baloney as a general guide, though it may mean something in terms of skikit-learn.
Hastie has a decent classification of the strengths and weaknesses of different learning algorithms in his book. It's not a decision tree. It will never be a decision tree.
Seeing this (a while ago) from the perspective of someone that was a complete noob to machine learning I found it extremely useful. It my not be the end-all-be-all, but I found it a useful graphic for figuring out which algorithms I should research to solve a specific ML problem.
I guess the chart is designed to those who are doing their first steps in the machine learning world. It makes order. BTW I have nothing to do with scikit-learn, I shared it as I found it to be useful.
Thanks for sharing it. I think it is a good start; it gives people something to react and build upon.
I don't see why the diagram has SVC and ensemble classifiers located in the "not working" path from KNeighbors Classifiers. It is reasonable to use an ensemble method independent of whether nearest neighbors works.
This looks very similar to the map described on dlib C++ Machine Learning library page. http://dlib.net/ml_guide.svg
I'd suggest giving Random Forests a call-out instead of leaving them hidden under ensemble methods in the diagram. I realize this is a clarity / detail tradeoff.
"This project has been temporarily blocked for exceeding its bandwidth threshold" I wonder why no one uses SourceForge anymore..
I feel like that's SourceForge's job. When something gets excitement, that's the last time you want to throw up a page like this. Poor user experience. Would, say, GitHub do this?
I suppose something else could be afoot, but I feel like controls should be in place for that elsewhere, e.g. you can't upload a gigantic binary.
What's going on here? The page is hosted on the scikit-learn.org domain, and I don't see any redirects or frames pointing to SourceForge. Or is SourceForge allowing projects to point arbitrary domains at it?
Yes it's hosted on SF with a CNAME dns entry. We want to move to github but we need to update the website build infra to do so.
dlib did something similar with different results.