14 KiB
clj-ml
A machine learning library for Clojure built on top of Weka and friends.
Installation
In order to install the library you must first install Leiningen.
To install from source
git clone the project, then run:
$ lein deps
$ lein javac
$ lein uberjar
Installing from Clojars
[cc.artifice/clj-ml "0.4.0-SNAPSHOT"]
Installing from Maven
(add Clojars repository)
<dependency>
<groupId>cc.artifice</groupId>
<artifactId>clj-ml</artifactId>
<version>0.4.0-SNAPSHOT</version>
</dependency>
Supported algorithms
-
Filters
- Discretization (supervised, unsupervised, PKI)
- Nominal to binary (supervised, unsupervised)
- Numeric to nominal
- String to word vector
- Attribute manipulation (reorder, add, remove range, remove percentage, etc.)
- Resample (supervised, unsupervised)
-
Classifiers
- k-Nearest neighbor
- Decision trees: C4.5/J4.8, Boosted stump, Random forest, Rotation forest, M5P
- Naive Bayes
- Multilayer perceptrons
- Support vector machines (grid-based training), SMO, Spegasos
-
Regression
- Linear
- Logistic
- Pace
- Additive gradient boosting
-
Clusterers
- k-Means
- Cobweb
- Expectation-maximization
Usage
API documenation can be found here.
I/O of data
user> (use 'clj-ml.io)
nil
user> (def ds (load-instances :arff "file:///home/josh/git/clj-ml/iris.arff"))
#'user/ds
user> ds
#<Instances @relation iris
@attribute sepallength numeric
@attribute sepalwidth numeric
@attribute petallength numeric
@attribute petalwidth numeric
@attribute class {Iris-setosa,Iris-versicolor,Iris-virginica}
@data
5.1,3.5,1.4,0.2,Iris-setosa
4.9,3,1.4,0.2,Iris-setosa
4.7,3.2,1.3,0.2,Iris-setosa
4.6,3.1,1.5,0.2,Iris-setosa
5,3.6,1.4,0.2,Iris-setosa
5.4,3.9,1.7,0.4,Iris-setosa
4.6,3.4,1.4,0.3,Iris-setosa
...
user> (def ds (load-instances :arff "http://repository.seasr.org/Datasets/UCI/arff/iris.arff"))
#'user/ds
user> (save-instances :csv "iris.csv" ds)
nil
user> (println (slurp "iris.csv"))
sepallength,sepalwidth,petallength,petalwidth,class
5.1,3.5,1.4,0.2,Iris-setosa
4.9,3,1.4,0.2,Iris-setosa
4.7,3.2,1.3,0.2,Iris-setosa
4.6,3.1,1.5,0.2,Iris-setosa
5,3.6,1.4,0.2,Iris-setosa
5.4,3.9,1.7,0.4,Iris-setosa
4.6,3.4,1.4,0.3,Iris-setosa
5,3.4,1.5,0.2,Iris-setosa
4.4,2.9,1.4,0.2,Iris-setosa
4.9,3.1,1.5,0.1,Iris-setosa
5.4,3.7,1.5,0.2,Iris-setosa
...
user> (def ds (load-instances :csv "file:///home/josh/git/clj-ml/iris.csv"))
#'user/ds
user> ds
#<Instances @relation stream
@attribute sepallength numeric
@attribute sepalwidth numeric
@attribute petallength numeric
@attribute petalwidth numeric
@attribute class {Iris-setosa,Iris-versicolor,Iris-virginica}
@data
5.1,3.5,1.4,0.2,Iris-setosa
4.9,3,1.4,0.2,Iris-setosa
4.7,3.2,1.3,0.2,Iris-setosa
4.6,3.1,1.5,0.2,Iris-setosa
5,3.6,1.4,0.2,Iris-setosa
5.4,3.9,1.7,0.4,Iris-setosa
4.6,3.4,1.4,0.3,Iris-setosa
5,3.4,1.5,0.2,Iris-setosa
Working with datasets
user> (use 'clj-ml.data)
nil
user> (def ds (make-dataset"my-name" [:length :width {:style nil} {:kind [:good :bad]}]
[[12 24 "longish" :good]
[8 5 "shortish" :bad]]))
#'user/ds
user> ds
#<ClojureInstances @relation my-name
@attribute length numeric
@attribute width numeric
@attribute style string
@attribute kind {good,bad}
@data
12,24,longish,good
8,5,shortish,bad>
user> (dataset-seq ds)
(#<Instance 12,24,longish,good> #<Instance 8,5,shortish,bad>)
user> (map instance-to-map (dataset-seq ds))
({:kind :good, :style "longish", :width 24.0, :length 12.0}
{:kind :bad, :style "shortish", :width 5.0, :length 8.0})
user> (map instance-to-vector (dataset-seq ds))
([12.0 24.0 "longish" :good] [8.0 5.0 "shortish" :bad])
Filtering datasets
user> (use 'clj-ml.filters 'clj-ml.io)
nil
user> (def ds (load-instances :csv "file:///home/josh/git/clj-ml/iris.csv"))
#'user/ds
user> (def discretize (make-filter :unsupervised-discretize
{:dataset-format ds
:attributes [:sepallength :petallength]}))
#'user/discretize
user> (def filtered-ds (filter-apply discretize ds))
#'user/filtered-ds
user> (map instance-to-map (dataset-seq filtered-ds))
({:class :Iris-setosa, :petalwidth 0.2, :petallength :'(-inf-1.59]',
:sepalwidth 3.5, :sepallength :'(5.02-5.38]'}
{:class :Iris-setosa, :petalwidth 0.2, :petallength :'(-inf-1.59]',
:sepalwidth 3.0, :sepallength :'(4.66-5.02]'}
{:class :Iris-setosa, :petalwidth 0.2, :petallength :'(-inf-1.59]',
:sepalwidth 3.2, :sepallength :'(4.66-5.02]'}
{:class :Iris-setosa, :petalwidth 0.2, :petallength :'(-inf-1.59]',
:sepalwidth 3.1, :sepallength :'(-inf-4.66]'}
{:class :Iris-setosa, :petalwidth 0.2, :petallength :'(-inf-1.59]',
:sepalwidth 3.6, :sepallength :'(4.66-5.02]'}
...) ;; the petallength and sepallength attributes are now nominal
Equivalently,
user> (def filtered-ds (->> "file:///home/josh/git/clj-ml/iris.csv"
(load-instances :csv)
(make-apply-filter :unsupervised-discretize
{:attributes [:sepallength :petallength]})))
Using classifiers
user> (use 'clj-ml.classifiers 'clj-ml.data 'clj-ml.utils)
nil
user> (def ds (-> (load-instances :arff "file:///home/josh/git/clj-ml/iris.arff")
(dataset-set-class :class)))
#'user/ds
user> (def classifier (-> (make-classifier :decision-tree :c45)
(classifier-train ds)))
#'user/classifier
user> (def instance (-> (first (dataset-seq ds))
(instance-set-class-missing)))
user> (classifier-classify classifier instance)
:Iris-setosa
Evaluation:
user> (def evaluation (classifier-evaluate classifier :cross-validation ds 10))
#'user/evaluation
user> (clojure.pprint/pprint (dissoc evaluation :summary :confusion-matrix))
{:incorrect 7.0,
:root-relative-squared-error 36.693518966642074,
:sf-entropy-gain -4076.3670930399717,
:recall
{:Iris-setosa 0.9795918367346939,
:Iris-versicolor 0.94,
:Iris-virginica 0.94},
:kb-information 217.7935138195151,
:kb-relative-information 13741.240800360849,
:false-positive-rate
{:Iris-setosa 0.0,
:Iris-versicolor 0.04040404040404041,
:Iris-virginica 0.030303030303030304},
:percentage-correct 95.30201342281879,
:roc-area
{:Iris-setosa 0.984845423317842,
:Iris-versicolor 0.9456,
:Iris-virginica 0.9496},
:kb-mean-information 1.4617014350303028,
:percentage-unclassified 0.0,
:percentage-incorrect 4.697986577181208,
:root-mean-squared-error 0.17297908222448935,
:unclassified 0.0,
:correlation-coefficient
{:nan "Can't compute correlation coefficient: class is nominal!"},
:correct 142.0,
:sf-mean-entropy-gain -27.358168409664238,
:mean-absolute-error 0.04083212821368881,
:relative-absolute-error 9.187228848079984,
:error-rate 0.04697986577181208,
:kappa 0.9295222650179066,
:f-measure
{:Iris-setosa 0.9896907216494846,
:Iris-versicolor 0.9306930693069307,
:Iris-virginica 0.94},
:false-negative-rate
{:Iris-setosa 0.02040816326530612,
:Iris-versicolor 0.06,
:Iris-virginica 0.06},
:evaluation-object #<Evaluation weka.classifiers.Evaluation@6a7272ca>,
:average-cost 0.0,
:precision
{:Iris-setosa 1.0,
:Iris-versicolor 0.9215686274509803,
:Iris-virginica 0.94}}
user> (println (:summary evaluation))
Correctly Classified Instances 142 95.302 %
Incorrectly Classified Instances 7 4.698 %
Kappa statistic 0.9295
Mean absolute error 0.0408
Root mean squared error 0.173
Relative absolute error 9.1872 %
Root relative squared error 36.6935 %
Total Number of Instances 149
Ignored Class Unknown Instances 1
nil
user> (println (:confusion-matrix evaluation))
=== Confusion Matrix ===
a b c <-- classified as
48 1 0 | a = Iris-setosa
0 47 3 | b = Iris-versicolor
0 3 47 | c = Iris-virginica
nil
Saving and restoring (trained) classifiers:
user> (serialize-to-file classifier "my-classifier.bin")
"my-classifier.bin"
user> (def classifier2 (deserialize-from-file "my-classifier.bin"))
#'user/classifier2
user> (classifier-classify classifier2 instance)
:Iris-setosa
Text document handling:
user> (def docs [{:title "Document title 1"
:fulltext "This is the fulltext..."
:terms {"Topic" ["Sports"]}}
{:title "Another document title"
:fulltext "Some more \"fulltext\"; rabbit artificial machine bananas"
:terms {"Topic" ["Politics" "Food"]}}])
#'user/docs
user> (docs-to-dataset docs "Topic" "Sports" 1 "/tmp" :stemmer true :lowercase false)
#<Instances @relation 'docs-weka.filters.unsupervised.attribute.StringToWordVector...'
@attribute class {no,yes}
@attribute title-1 numeric
@attribute title-Another numeric
@attribute title-Document numeric
@attribute title-document numeric
@attribute title-titl numeric
@attribute fulltext-Some numeric
@attribute fulltext-This numeric
@attribute fulltext-artifici numeric
@attribute fulltext-banana numeric
@attribute fulltext-fulltext numeric
@attribute fulltext-is numeric
@attribute fulltext-machin numeric
@attribute fulltext-more numeric
@attribute fulltext-rabbit numeric
@attribute fulltext-the numeric
@data
{0 yes,1 0.480453,3 0.480453,7 0.480453,11 0.480453,15 0.480453}
{2 0.480453,4 0.480453,6 0.480453,8 0.480453,9 0.480453,12 0.480453,13 0.480453,14 0.480453}>
user>
Words appearing in the dataset will only be those appearing in the documents (or a subset; by default, the most common 1000 words). This presents a problem when new documents are loaded and used in a classifier trained on other documents. The classifier will not know how to handle word attributes that are not present in the training set.
The docs-to-dataset
function provides the ability to save the
training documents dataset and "filter" the testing documents through
this dataset to ensure the same word attributes are extracted for both
sets. The following example shows that the words "foo, bar, baz, quux"
are ignored in the new (testing) documents, and all the original
attributes in the training dataset are retained.
user> (docs-to-dataset docs "Topic" "Sports" 1 "/tmp"
:stemmer true :lowercase false :training true)
#<Instances @relation 'docs-weka.filters.unsupervised.attribute.StringToWordVector...'
@attribute class {no,yes}
@attribute title-1 numeric
@attribute title-Another numeric
@attribute title-Document numeric
@attribute title-document numeric
@attribute title-titl numeric
@attribute fulltext-Some numeric
@attribute fulltext-This numeric
@attribute fulltext-artifici numeric
@attribute fulltext-banana numeric
@attribute fulltext-fulltext numeric
@attribute fulltext-is numeric
@attribute fulltext-machin numeric
@attribute fulltext-more numeric
@attribute fulltext-rabbit numeric
@attribute fulltext-the numeric
@data
{2 0.480453,4 0.480453,6 0.480453,8 0.480453,9 0.480453,12 0.480453,13 0.480453,14 0.480453}
{0 yes,1 0.480453,3 0.480453,7 0.480453,11 0.480453,15 0.480453}>
user> (def docs2 [{:title "Document title 1 foo bar"
:fulltext "baz rabbit quux"
:terms {"Topic" ["Sports"]}}])
#'user/docs2
user> (docs-to-dataset docs2 "Topic" "Sports" 1 "/tmp"
:stemmer true :lowercase false :testing true)
#<Instances @relation 'docs-weka.filters.unsupervised.attribute.StringToWordVector...'
@attribute class {no,yes}
@attribute title-1 numeric
@attribute title-Another numeric
@attribute title-Document numeric
@attribute title-document numeric
@attribute title-titl numeric
@attribute fulltext-Some numeric
@attribute fulltext-This numeric
@attribute fulltext-artifici numeric
@attribute fulltext-banana numeric
@attribute fulltext-fulltext numeric
@attribute fulltext-is numeric
@attribute fulltext-machin numeric
@attribute fulltext-more numeric
@attribute fulltext-rabbit numeric
@attribute fulltext-the numeric
@data
{0 yes,1 0.480453,3 0.480453,14 0.480453}>
user>
Using clusterers
user> (use 'clj-ml.clusterers)
nil
user> (def ds (-> (load-instances :arff "file:///home/josh/git/clj-ml/iris.arff")
(dataset-remove-attribute-at 4)))
#'user/ds
user> ds
#<Instances @relation iris
@attribute sepallength numeric
@attribute sepalwidth numeric
@attribute petallength numeric
@attribute petalwidth numeric
@data
5.1,3.5,1.4,0.2
4.9,3,1.4,0.2
4.7,3.2,1.3,0.2
4.6,3.1,1.5,0.2
5,3.6,1.4,0.2
5.4,3.9,1.7,0.4
4.6,3.4,1.4,0.3
...
user> (def clusterer (make-clusterer :k-means {:number-clusters 3}))
#'user/clusterer
user> (clusterer-build clusterer ds)
nil
user> clusterer
#<SimpleKMeans
kMeans
======
Number of iterations: 6
Within cluster sum of squared errors: 6.998114004826762
Missing values globally replaced with mean/mode
Cluster centroids:
Cluster#
Attribute Full Data 0 1 2
(150) (61) (50) (39)
=========================================================
sepallength 5.8433 5.8885 5.006 6.8462
sepalwidth 3.054 2.7377 3.418 3.0821
petallength 3.7587 4.3967 1.464 5.7026
petalwidth 1.1987 1.418 0.244 2.0795
>
user>
Thanks YourKit!
YourKit is kindly supporting open source projects with its full-featured Java Profiler. YourKit, LLC is the creator of innovative and intelligent tools for profiling Java and .NET applications. Take a look at YourKit's leading software products: YourKit Java Profiler and YourKit .NET Profiler.
License
MIT License