In [1]:

    

%cd ~/NetBeansProjects/ExpLosion/
from notebooks.common_imports import *
from gui.models import get_ci

sns.timeseries.algo.bootstrap = my_bootstrap
sns.categorical.bootstrap = my_bootstrap


    

    




/Users/miroslavbatchkarov/NetBeansProjects/ExpLosion

In [2]:

    

d = {'labelled': 'reuters21578/r8-tagged-grouped',
    'expansions__noise': 0,
     'expansions__vectors__composer': 'Add',
     'expansions__vectors__rep': 0, 
    'expansions__vectors__avg':0,
     'expansions__vectors__unlabelled_percentage': 100}


    

In [3]:

    

ids = Experiment.objects.filter(**d).values_list('id', flat=True)
print('experiments are', ids, len(ids))
df = dataframe_from_exp_ids(ids, {'id':'id',
                                    'unlab':'expansions__vectors__unlabelled',
                                 'algo':'expansions__vectors__algorithm',
                                 'k':'expansions__k'}, abbreviate=False)


    

    




experiments are [135, 392, 403, 414, 415, 416, 417, 418, 419, 420] 10
algo has 5000 values
folds has 5000 values
id has 5000 values
k has 5000 values
unlab has 5000 values
Accuracy has 5000 values

In [4]:

    

performance_table(df)


    

    




keeping {'algo', 'k', 'unlab'}






    
Out[4]:






  

    

      

      

      

      
mean
      
ci_width
    
    

      
algo
      
k
      
unlab
      

      

    
  
  

    

      
count_dependencies
      
3
      
gigaw
      
64.153577
      
3.495008
    
    

      
wiki
      
61.465959
      
3.677786
    
    

      
count_windows
      
3
      
gigaw
      
71.938194
      
3.168837
    
    

      
wiki
      
66.948773
      
3.719709
    
    

      
glove
      
3
      
wiki
      
61.848222
      
2.969950
    
    

      
word2vec
      
3
      
cwiki
      
64.885893
      
3.370298
    
    

      
gigaw
      
74.164143
      
2.981941
    
    

      
wiki
      
68.037037
      
3.223113
    
    

      
30
      
wiki
      
59.471400
      
2.998158
    
    

      
60
      
wiki
      
56.984646
      
3.127234
    
  

In [5]:

    

sns.factorplot(data=df[df.k=='3'], y='Accuracy', x='unlab', hue='algo',
               ci=100, kind='bar');


    

In [6]:

    

sign_df = get_demsar_params(ids, ['expansions__vectors__unlabelled', 
                                  'expansions__vectors__algorithm', 
                                  'expansions__k'])[0]
sign_df[(sign_df.significant==True)&(sign_df.name2=='gigaw-w2v-3')].sort('mean_diff')


    

    
Out[6]:






  

    

      

      
name1
      
acc1
      
name2
      
acc2
      
mean_diff
      
pval
      
significant
    
  
  

    

      
6
      
wiki-w2v-3
      
0.680912
      
gigaw-w2v-3
      
0.741514
      
0.060603
      
0
      
True
    
    

      
32
      
wiki-W-3
      
0.669479
      
gigaw-w2v-3
      
0.741514
      
0.072036
      
0
      
True
    
    

      
39
      
cwiki-w2v-3
      
0.648903
      
gigaw-w2v-3
      
0.741514
      
0.092612
      
0
      
True
    
    

      
27
      
wiki-glove-3
      
0.618065
      
gigaw-w2v-3
      
0.741514
      
0.123449
      
0
      
True
    
    

      
36
      
wiki-D-3
      
0.615618
      
gigaw-w2v-3
      
0.741514
      
0.125896
      
0
      
True
    
    

      
14
      
wiki-w2v-30
      
0.594776
      
gigaw-w2v-3
      
0.741514
      
0.146739
      
0
      
True
    
    

      
21
      
wiki-w2v-60
      
0.569159
      
gigaw-w2v-3
      
0.741514
      
0.172355
      
0
      
True
    
  

In [7]:

    

for i in ids:
    cv_score = Results.objects.get(id=i, classifier=CLASSIFIER).accuracy_mean
    bs_score = get_ci(i)[0]
    print((cv_score - bs_score)*100)


    

    




0.0541310541311
0.00616208975218
-0.0686962276638
-0.0416785206259
-0.000920245398772
0.0958301743745
0.00438871473355
-0.0127067014795
0.0236111111111
-0.0104825291181

In [ ]: