

In [1]:

    
%autosave 0









    














    



Autosave disabled

suneku.classifier



In [2]:

    
from suneku import hello
from suneku.classifier import LogisticClassifier, traintest
hello(LogisticClassifier)









    



type

    Classify rows of a matrix into one of several categories.
    Return a DataFrame with category and probability for each row.

    Initialize using training data with known categories.
    Call with new data to classify each row.

    Input
        data    any valid DataFrame input: Training data.
        target  string or int: Name of column with known classes.
        **kwargs are passed to scikit-learn model.

demo



In [3]:

    
demo = LogisticClassifier.demo()
print(demo)









    



LogisticClassifier
target  : species
samples : 150
model   : LogisticRegressionCV
features:
  sepal length (cm)
  sepal width (cm)
  petal length (cm)
  petal width (cm)
classes :
  setosa
  versicolor
  virginica



In [4]:

    
demo.data.head()









    Out[4]:







  
    
      
      species
      sepal length (cm)
      sepal width (cm)
      petal length (cm)
      petal width (cm)
    
  
  
    
      0
      setosa
      5.1
      3.5
      1.4
      0.2
    
    
      1
      setosa
      4.9
      3.0
      1.4
      0.2
    
    
      2
      setosa
      4.7
      3.2
      1.3
      0.2
    
    
      3
      setosa
      4.6
      3.1
      1.5
      0.2
    
    
      4
      setosa
      5.0
      3.6
      1.4
      0.2

partition rows into training and testing



In [5]:

    
traindata,testdata = traintest(demo.data,0.33)
traindata.sample(5)









    



Train with 49 of 150 rows






    Out[5]:







  
    
      
      species
      sepal length (cm)
      sepal width (cm)
      petal length (cm)
      petal width (cm)
    
  
  
    
      80
      versicolor
      5.5
      2.4
      3.8
      1.1
    
    
      50
      versicolor
      7.0
      3.2
      4.7
      1.4
    
    
      95
      versicolor
      5.7
      3.0
      4.2
      1.2
    
    
      114
      virginica
      5.8
      2.8
      5.1
      2.4
    
    
      81
      versicolor
      5.5
      2.4
      3.7
      1.0

build a LogisticClassifier

Pre-processing and learning are automatic.



In [6]:

    
params = {'penalty':'l1','solver':'liblinear'}
classifier = LogisticClassifier(traindata,'species',**params)
print(classifier)









    



LogisticClassifier
target  : species
samples : 49
model   : LogisticRegressionCV
features:
  sepal length (cm)
  sepal width (cm)
  petal length (cm)
  petal width (cm)
classes :
  setosa
  versicolor
  virginica

inspect data



In [7]:

    
pairgrid = classifier.show(palette=list('rgb'))

inspect model



In [8]:

    
print(classifier.model)









    



LogisticRegressionCV(Cs=10, class_weight=None, cv=None, dual=False,
           fit_intercept=True, intercept_scaling=1.0, max_iter=100,
           multi_class='ovr', n_jobs=1, penalty='l1', random_state=None,
           refit=True, scoring=None, solver='liblinear', tol=0.0001,
           verbose=0)



In [9]:

    
classifier.coefs()









    Out[9]:







  
    
      
      sepal length (cm)
      sepal width (cm)
      petal length (cm)
      petal width (cm)
    
  
  
    
      setosa
      0.000000
      0.433493
      -2.110181
      0.000000
    
    
      versicolor
      1.209620
      -1.842864
      0.000000
      -1.452936
    
    
      virginica
      -10.225122
      1.321914
      46.739045
      37.837794

classify things



In [10]:

    
classifier(testdata).sample(5)









    Out[10]:







  
    
      
      class
      p_class
    
  
  
    
      51
      versicolor
      0.508176
    
    
      47
      setosa
      0.852205
    
    
      42
      setosa
      0.880945
    
    
      101
      virginica
      0.705137
    
    
      145
      virginica
      0.756979



In [11]:

    
classifier.probs(testdata).sample(5)









    Out[11]:







  
    
      
      setosa
      versicolor
      virginica
    
  
  
    
      44
      0.973617
      0.026383
      2.518699e-45
    
    
      149
      0.085775
      0.132294
      7.819317e-01
    
    
      8
      0.732103
      0.267897
      1.651063e-51
    
    
      89
      0.233305
      0.766695
      3.377614e-12
    
    
      82
      0.293780
      0.706220
      3.385790e-16

test with known classes



In [12]:

    
print('Top 5 worst predictions:')
classifier.test(testdata).nlargest(5,'surprise')









    



Top 5 worst predictions:
89.1% correct

class       setosa  versicolor  virginica
species                                  
setosa          34           1          0
versicolor       5          26          3
virginica        0           2         30







    Out[12]:







  
    
      
      class
      correct
      p_class
      reality
      surprise
    
  
  
    
      119
      versicolor
      False
      0.839290
      virginica
      2.637469
    
    
      85
      setosa
      False
      0.799135
      versicolor
      2.315703
    
    
      70
      virginica
      False
      0.797825
      versicolor
      2.306324
    
    
      133
      versicolor
      False
      0.671909
      virginica
      1.607831
    
    
      83
      virginica
      False
      0.626219
      versicolor
      1.419734

try again with new parameters



In [13]:

    
params = dict()
params['multi_class'] = 'multinomial'
params['penalty'] = 'l2'
params['solver'] = 'lbfgs'

classifier.learn(**params)
classifier.coefs()









    Out[13]:







  
    
      
      sepal length (cm)
      sepal width (cm)
      petal length (cm)
      petal width (cm)
    
  
  
    
      setosa
      -2.335692
      3.216301
      -5.930037
      -5.156432
    
    
      versicolor
      2.990490
      -1.512983
      -3.871019
      -3.392405
    
    
      virginica
      -0.654797
      -1.703318
      9.801057
      8.548838



In [14]:

    
print('Top 5 worst predictions:')
classifier.test(testdata).nlargest(5,'surprise')









    



Top 5 worst predictions:
97.0% correct

class       setosa  versicolor  virginica
species                                  
setosa          35           0          0
versicolor       0          32          2
virginica        0           1         31







    Out[14]:







  
    
      
      class
      correct
      p_class
      reality
      surprise
    
  
  
    
      70
      virginica
      False
      0.918401
      versicolor
      3.615300
    
    
      83
      virginica
      False
      0.850194
      versicolor
      2.738833
    
    
      133
      versicolor
      False
      0.690307
      virginica
      1.691091
    
    
      119
      virginica
      True
      0.501302
      virginica
      0.996248
    
    
      41
      setosa
      True
      0.528061
      setosa
      0.921225

	species	sepal length (cm)	sepal width (cm)	petal length (cm)	petal width (cm)
0	setosa	5.1	3.5	1.4	0.2
1	setosa	4.9	3.0	1.4	0.2
2	setosa	4.7	3.2	1.3	0.2
3	setosa	4.6	3.1	1.5	0.2
4	setosa	5.0	3.6	1.4	0.2

	species	sepal length (cm)	sepal width (cm)	petal length (cm)	petal width (cm)
80	versicolor	5.5	2.4	3.8	1.1
50	versicolor	7.0	3.2	4.7	1.4
95	versicolor	5.7	3.0	4.2	1.2
114	virginica	5.8	2.8	5.1	2.4
81	versicolor	5.5	2.4	3.7	1.0

	sepal length (cm)	sepal width (cm)	petal length (cm)	petal width (cm)
setosa	0.000000	0.433493	-2.110181	0.000000
versicolor	1.209620	-1.842864	0.000000	-1.452936
virginica	-10.225122	1.321914	46.739045	37.837794

	class	p_class
51	versicolor	0.508176
47	setosa	0.852205
42	setosa	0.880945
101	virginica	0.705137
145	virginica	0.756979

	setosa	versicolor	virginica
44	0.973617	0.026383	2.518699e-45
149	0.085775	0.132294	7.819317e-01
8	0.732103	0.267897	1.651063e-51
89	0.233305	0.766695	3.377614e-12
82	0.293780	0.706220	3.385790e-16

	class	correct	p_class	reality	surprise
119	versicolor	False	0.839290	virginica	2.637469
85	setosa	False	0.799135	versicolor	2.315703
70	virginica	False	0.797825	versicolor	2.306324
133	versicolor	False	0.671909	virginica	1.607831
83	virginica	False	0.626219	versicolor	1.419734

	sepal length (cm)	sepal width (cm)	petal length (cm)	petal width (cm)
setosa	-2.335692	3.216301	-5.930037	-5.156432
versicolor	2.990490	-1.512983	-3.871019	-3.392405
virginica	-0.654797	-1.703318	9.801057	8.548838

	class	correct	p_class	reality	surprise
70	virginica	False	0.918401	versicolor	3.615300
83	virginica	False	0.850194	versicolor	2.738833
133	versicolor	False	0.690307	virginica	1.691091
119	virginica	True	0.501302	virginica	0.996248
41	setosa	True	0.528061	setosa	0.921225