5 ms·

Machine Learning Cheat Sheet Map

114 points by cognibits 12y ago

dj-wonk 12y ago
Just 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.linear_model.LogisticRegression.html#sklearn.linear_model.LogisticRegression http://scikit-learn.org/stable/modules/generated/sklearn.lin...
dj-wonk 12y ago
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.
dj-wonk 12y ago
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.
- cognibits 12y ago
  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.
  - dj-wonk 12y ago
    Thanks for sharing it. I think it is a good start; it gives people something to react and build upon.
- yaur 12y ago
  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.
- scottlocklin 12y ago
  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.
dj-wonk 12y ago
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.
dj-wonk 12y ago
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. :)
iLoch 12y ago
"This project has been temporarily blocked for exceeding its bandwidth threshold" I wonder why no one uses SourceForge anymore..
- CyberShadow 12y ago
  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?
  - ogrisel 12y ago
    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.
- factotvm 12y ago
  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.
ColinWright 12y ago
It's currently down - is it the same one discussed at some length a year ago? https://news.ycombinator.com/item?id=5831512 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://eferm.com/wp-content/uploads/2011/05/cheat3.pdf http://peekaboo-vision.blogspot.ca/2013/01/machine-learning-cheat-sheet-for-scikit.html http://peekaboo-vision.blogspot.ca/2013/01/machine-learning-... http://rise.cse.iitm.ac.in/wiki/index.php/Introduction_to_Machine_Learning http://rise.cse.iitm.ac.in/wiki/index.php/Introduction_to_Ma... http://mlg.eng.cam.ac.uk/creed/Notes/ML_Compendium.pdf http://mlg.eng.cam.ac.uk/creed/Notes/ML_Compendium.pdf
lazycrazyowl 12y ago
This looks very similar to the map described on dlib C++ Machine Learning library page. http://dlib.net/ml_guide.svg http://dlib.net/ml_guide.svg
Pamar 12y ago
"Mathbabe" is currently working on something similar: http://mathbabe.org/2014/08/28/a-decision-tree-for-decision-trees/ 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).
a8da6b0c91d 12y ago
dlib did something similar with different results. http://dlib.net/ml_guide.svg http://dlib.net/ml_guide.svg

114 points by cognibits 12y ago

Just 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.linear_model.LogisticRegression.html#sklearn.linear_model.LogisticRegression http://scikit-learn.org/stable/modules/generated/sklearn.lin...

dj-wonk 12y ago

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.

dj-wonk 12y ago

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.

cognibits 12y ago
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.
- dj-wonk 12y ago
  Thanks for sharing it. I think it is a good start; it gives people something to react and build upon.
yaur 12y ago
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.
scottlocklin 12y ago
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.

dj-wonk 12y ago

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.

dj-wonk 12y ago

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. :)

iLoch 12y ago

"This project has been temporarily blocked for exceeding its bandwidth threshold" I wonder why no one uses SourceForge anymore..

CyberShadow 12y ago
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?
- ogrisel 12y ago
  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.
factotvm 12y ago
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.

ColinWright 12y ago

It's currently down - is it the same one discussed at some length a year ago? https://news.ycombinator.com/item?id=5831512 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://eferm.com/wp-content/uploads/2011/05/cheat3.pdf http://peekaboo-vision.blogspot.ca/2013/01/machine-learning-cheat-sheet-for-scikit.html http://peekaboo-vision.blogspot.ca/2013/01/machine-learning-... http://rise.cse.iitm.ac.in/wiki/index.php/Introduction_to_Machine_Learning http://rise.cse.iitm.ac.in/wiki/index.php/Introduction_to_Ma... http://mlg.eng.cam.ac.uk/creed/Notes/ML_Compendium.pdf http://mlg.eng.cam.ac.uk/creed/Notes/ML_Compendium.pdf

lazycrazyowl 12y ago

This looks very similar to the map described on dlib C++ Machine Learning library page. http://dlib.net/ml_guide.svg http://dlib.net/ml_guide.svg

Pamar 12y ago

"Mathbabe" is currently working on something similar: http://mathbabe.org/2014/08/28/a-decision-tree-for-decision-trees/ 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).

a8da6b0c91d 12y ago

dlib did something similar with different results. http://dlib.net/ml_guide.svg http://dlib.net/ml_guide.svg