

In [1]:

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

from gui.user_code import get_demsar_diagram, get_demsar_params
from gui.output_utils import *


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









    



/Users/miroslavbatchkarov/NetBeansProjects/ExpLosion

Number of clusters for different composers and unlabelled corpora



In [2]:

    
query_dict = {
             'clusters__noise': 0,
             'clusters__vectors__algorithm__in': ['turian'],
             'clusters__vectors__composer__in': ['Add', 'Socher'],
             'clusters__vectors__unlabelled__in': ['turian'],
             'clusters__vectors__dimensionality': 100,
             'clusters__vectors__rep': 0,
             'clusters__vectors__unlabelled_percentage': 100,
             'document_features_ev': 'AN+NN',
             'document_features_tr': 'J+N+AN+NN',
             'labelled': 'amazon_grouped-tagged'}

def go(query_dict):
    ids = Experiment.objects.filter(**query_dict).values_list('id', flat=True)
    print(ids)
    df = dataframe_from_exp_ids(ids, {'Clusters':'clusters__num_clusters', 
                                      'Composer':'clusters__vectors__composer',
                                      'Unlabelled':'clusters__vectors__unlabelled'})
    df = df.convert_objects(convert_numeric=True)
    g = sns.tsplot(data=df, time='Clusters', condition='Composer', 
               value='Accuracy', unit='folds',
               marker='s', linewidth=4, ci=68) 
    g.legend(loc='lower right')
    g.set(ylim=(0.45, .8), xlim=(0, 2000))

# g.set(xscale='log')
# sns.despine(left=True)



In [3]:

    
with sns.color_palette("cubehelix", 4):
    go(query_dict)
# plt.savefig('plot-clusters1.pdf', format='pdf', dpi=300, bbox_inches='tight', pad_inches=0.1)









    



[240, 241, 242, 243, 244, 245, 246, 247, 248, 249]
Unlabelled has 5000 values
folds has 5000 values
Accuracy has 5000 values
Clusters has 5000 values
Composer has 5000 values



In [4]:

    
query_dict['clusters__vectors__algorithm__in'] = ['word2vec']
query_dict['clusters__vectors__composer__in'] = ['Add', 'Mult']
query_dict['clusters__vectors__unlabelled__in'] = ['gigaw']

with sns.color_palette("cubehelix", 4):
    go(query_dict)
# plt.savefig('plot-clusters2.pdf', format='pdf', dpi=300, bbox_inches='tight', pad_inches=0.1)









    



[250, 251, 252, 253, 254, 260, 261, 262, 263, 264]
Unlabelled has 5000 values
folds has 5000 values
Accuracy has 5000 values
Clusters has 5000 values
Composer has 5000 values



In [5]:

    
query_dict['clusters__vectors__unlabelled__in'] = ['wiki']
with sns.color_palette("cubehelix", 4):
    go(query_dict)
plt.savefig('plot-clusters3.pdf', format='pdf', dpi=300, bbox_inches='tight', pad_inches=0.1)









    



/Users/miroslavbatchkarov/anaconda/envs/py3k/lib/python3.3/site-packages/matplotlib/backends/backend_pdf.py:2184: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
  different = bool(ours != theirs)

WARNING:py.warnings:/Users/miroslavbatchkarov/anaconda/envs/py3k/lib/python3.3/site-packages/matplotlib/backends/backend_pdf.py:2184: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
  different = bool(ours != theirs)







    



[255, 256, 257, 258, 259, 265, 266, 267, 268, 269]
Unlabelled has 5000 values
folds has 5000 values
Accuracy has 5000 values
Clusters has 5000 values
Composer has 5000 values



In [6]:

    
query_dict = {
'clusters__num_clusters': 100,
'clusters__vectors__algorithm': 'word2vec',
'clusters__vectors__composer': 'Add',
'clusters__vectors__dimensionality': 100,
'clusters__vectors__rep': 0,
'clusters__vectors__unlabelled': 'wiki',
# 'clusters__vectors__unlabelled_percentage': 70,
'document_features_ev': 'AN+NN',
'document_features_tr': 'J+N+AN+NN',
# 'labelled': 'amazon_grouped-tagged'
}
ids = Experiment.objects.filter(**query_dict).values_list('id', flat=True)
print(ids)
df = dataframe_from_exp_ids(ids, {'Labelled':'labelled', 
                                  'Percent':'clusters__vectors__unlabelled_percentage'})
df = df.convert_objects(convert_numeric=True)
df.Labelled[df.Labelled == 'AM'] = 'Amazon'
with sns.color_palette("cubehelix", 4):
    g = sns.tsplot(data=df, time='Percent', condition='Labelled', 
               value='Accuracy', unit='folds',
               marker='s', linewidth=4, ci=68);
g.legend(loc='lower right')
plt.savefig('plot-clusters4.pdf', format='pdf', dpi=300, bbox_inches='tight', pad_inches=0.1)









    



/Users/miroslavbatchkarov/anaconda/envs/py3k/lib/python3.3/site-packages/IPython/kernel/__main__.py:18: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy

WARNING:py.warnings:/Users/miroslavbatchkarov/anaconda/envs/py3k/lib/python3.3/site-packages/IPython/kernel/__main__.py:18: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy







    



[265, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295]
folds has 5500 values
Accuracy has 5500 values
Labelled has 5500 values
Percent has 5500 values

Ideally I want something like this, but seaborn is buggy



In [7]:

    
query_dict = {
             'clusters__noise':0,
             'clusters__vectors__algorithm__in': ['turian', 'word2vec'],
             'clusters__vectors__composer__in': ['Add', 'Socher', 'Mult'],
             'clusters__vectors__unlabelled__in': ['turian', 'wiki', 'gigaw'],
             'clusters__vectors__dimensionality': 100,
             'clusters__vectors__rep': 0,
             'clusters__vectors__unlabelled_percentage': 100,
             'document_features_ev': 'AN+NN',
             'document_features_tr': 'J+N+AN+NN',
             'labelled': 'amazon_grouped-tagged'}
ids = Experiment.objects.filter(**query_dict).values_list('id', flat=True)
print(ids)
df = dataframe_from_exp_ids(ids, {'Clusters':'clusters__num_clusters', 
                                  'Composer':'clusters__vectors__composer',
                                  'Unlabelled':'clusters__vectors__unlabelled'})
df = df.convert_objects(convert_numeric=True)
g = sns.FacetGrid(df, col='Unlabelled')  
g.map_dataframe(tsplot_for_facetgrid, time='Clusters', condition='Composer', 
                   value='Accuracy', unit='folds',
                   marker='s', linewidth=4, ci=68);









    



[240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255, 256, 257, 258, 259, '...(remaining elements truncated)...']
Unlabelled has 15000 values
folds has 15000 values
Accuracy has 15000 values
Clusters has 15000 values
Composer has 15000 values

Noise validation



In [8]:

    
d = {#'clusters__noise': 2.0,
     'clusters__num_clusters': 100,
     'clusters__vectors__algorithm': 'word2vec',
     'clusters__vectors__avg': 0,
     'clusters__vectors__composer': 'Add',
     'clusters__vectors__dimensionality': 100,
     'clusters__vectors__reorder': 0,
     'clusters__vectors__rep': 0,
     'clusters__vectors__unlabelled': 'gigaw',
     'clusters__vectors__unlabelled_percentage': 100,
     'document_features_ev': 'AN+NN',
     'document_features_tr': 'J+N+AN+NN',
     'labelled': 'amazon_grouped-tagged'}
ids = Experiment.objects.filter(**d).values_list('id', flat=True)
df = dataframe_from_exp_ids(ids,
                           {'noise':'clusters__noise',
                           'id':'id'}).convert_objects(convert_numeric=True)
print(ids)
sns.tsplot(df, time='noise',value='Accuracy',unit='folds')
plt.savefig('plot-clustered-noise-validation.pdf', format='pdf', dpi=300, bbox_inches='tight', pad_inches=0.1);









    



noise has 5500 values
folds has 5500 values
Accuracy has 5500 values
id has 5500 values
[260, 373, 374, 375, 376, 377, 378, 379, 380, 381, 382]

Other experiments that do not merit a graph



In [9]:

    
from gui.user_code import pairwise_significance_exp_ids, get_data_for_signif_test
ids = Experiment.objects.filter(clusters__noise=0.0,
                                labelled='amazon_grouped-tagged',
                                clusters__vectors__composer='Add',
                                clusters__num_clusters=100,
                                clusters__vectors__unlabelled_percentage=100).values_list('id', flat=True)
ids









    Out[9]:





[245, 260, 265, 371, 372]



In [10]:

    
df = dataframe_from_exp_ids(ids, {'algo':'clusters__vectors__algorithm',
                                  'corpus':'clusters__vectors__unlabelled',
                                  'id':'id'})
df = df.convert_objects(convert_numeric=True)
df.head().dtypes









    



corpus has 2500 values
folds has 2500 values
Accuracy has 2500 values
algo has 2500 values
id has 2500 values






    Out[10]:





Accuracy    float64
algo         object
corpus       object
folds         int64
id            int64
dtype: object



In [11]:

    
df.groupby(['algo', 'corpus']).mean()



In [12]:

    
for i in ids:
    low, mid, high = get_ci(i)[:-1]
    print(i, (high-low)/2)









    



245 0.00523332688962
260 0.00493037365177
265 0.00494512195122
371 0.00486921111466
372 0.00438864057212



In [ ]:

		Accuracy	folds	id
algo	corpus
glove	wiki	0.695488	249.5	372
turian	turian	0.483898	249.5	245
w2v	cwiki	0.674422	249.5	371
	gigaw	0.553538	249.5	260
	wiki	0.656727	249.5	265