Feature Request: Apriori algorithm (working code)

[dfrusdn]

I would like to contribute my Apriori algorithm found here:

http://codereview.stackexchange.com/questions/38101/optimizing-apriori-algorithm-python-pandas

This algorithm is useful for pattern mining.

It does not have all the needed features, but would be a good point to start.

[GaelVaroquaux]

Hi

Thanks a lot for your enthousiasme. What's the relevant publication. Algorithms included in scikit-learn must be literature standards and come from publications with a high citation count. 

Cheers

Gaël

-------- Original message --------

From: dfrusdn notifications@github.com

Date:19/02/2014 04:01 (GMT+01:00)

To: scikit-learn/scikit-learn scikit-learn@noreply.github.com

Subject: [scikit-learn] Feature Request: Apriori algorithm (working code) (#2872)

I would like to contribute my Apriori algorithm found here:

http://codereview.stackexchange.com/questions/38101/optimizing-apriori-algorithm-python-pandas

This algorithm is useful for pattern mining.

It does not have all the needed features, but would be a good point to start.

—
Reply to this email directly or view it on GitHub.

[jnothman]

Apriori has no shortage of pedigree in data mining (
https://en.wikipedia.org/wiki/Apriori_algorithm), but whether it's relevant
to scikit-learn is another matter.

On 19 February 2014 16:57, Gael Varoquaux notifications@github.com wrote:

Hi

Thanks a lot for your enthousiasme. What's the relevant publication.
Algorithms included in scikit-learn must be literature standards and come
from publications with a high citation count.

Cheers

Gaël

-------- Original message --------

From: dfrusdn < notifications@github.com>

Date:19/02/2014 04:01 (GMT+01:00)

To: scikit-learn/scikit-learn scikit-learn@noreply.github.com

Subject: [scikit-learn] Feature Request: Apriori algorithm (working code) (#2872)

I would like to contribute my Apriori algorithm found here:

http://codereview.stackexchange.com/questions/38101/optimizing-apriori-algorithm-python-pandas

This algorithm is useful for pattern mining.

It does not have all the needed features, but would be a good point to
start.

Reply to this email directly or view it on GitHub.

Reply to this email directly or view it on GitHubhttps://github.com//issues/2872#issuecomment-35468796
.

[ccsv]

@jnothman
I think this belongs in Scikit-learn it is a precursor to the CBA algorithm
http://cgi.csc.liv.ac.uk/~frans/KDD/Software/CBA/cba.html

Scikit-learns front page states it provides "Simple and efficient tools for data mining and data analysis"

Associative rule learning is a good addition to the data mining component
http://en.wikipedia.org/wiki/Association_rule_learning

The problems with that algorithm would be the for loops for large data sets and there should be a calculation for confidence for each pattern.

[GaelVaroquaux]

Ok I wasn't completely sure that we were talking of this one. 

Gael

-------- Original message --------

From: jnothman notifications@github.com

Date:19/02/2014 07:10 (GMT+01:00)

To: scikit-learn/scikit-learn scikit-learn@noreply.github.com

Cc: Gael Varoquaux gael.varoquaux@normalesup.org

Subject: Re: [scikit-learn] Feature Request: Apriori algorithm (working code) (#2872)

Apriori has no shortage of pedigree in data mining ( https://en.wikipedia.org/wiki/Apriori_algorithm), but whether it's relevant to scikit-learn is another matter.

On 19 February 2014 16:57, Gael Varoquaux notifications@github.com wrote:

Hi

Thanks a lot for your enthousiasme. What's the relevant publication.
Algorithms included in scikit-learn must be literature standards and come
from publications with a high citation count.

Cheers

Gaël

-------- Original message --------

From: dfrusdn < notifications@github.com>

Date:19/02/2014 04:01 (GMT+01:00)

To: scikit-learn/scikit-learn scikit-learn@noreply.github.com

Subject: [scikit-learn] Feature Request: Apriori algorithm (working code) (#2872)

I would like to contribute my Apriori algorithm found here:

http://codereview.stackexchange.com/questions/38101/optimizing-apriori-algorithm-python-pandas

This algorithm is useful for pattern mining.

It does not have all the needed features, but would be a good point to
start.

Reply to this email directly or view it on GitHub.

Reply to this email directly or view it on GitHubhttps://github.com//issues/2872#issuecomment-35468796
.

—
Reply to this email directly or view it on GitHub.

[larsmans]

A pandas dependency is not acceptable; I for one cannot read the code you posted. What will be the interface for this algorithm? I'm not convinced that it can fit the existing API unless it's presented as a kernel approximation algorithm.

[jnothman]

I think it only fits in the context of something like CBA which classifies documents by learning association rules that map feature groups to target labels. In that context, the learnt rule-set constitutes the model... but not all rules in the general apriori are utilised.

[larsmans]

#2662 is a generalization of this feature request, closing.