The Curator

Blog Posts

• Can Word2Vec Describe Art-Historical Categories?
• Abusing Word2Vec, Part II
• Calculating Error Rates: The Curator

Benchmarks

 version      mean    stddev  epochs  size  sg train time  vocab  labeled  topn    
   1.8.2  0.731169  0.400759      10   100   1      1337s   1011      415     2     
   1.9.2  0.757323  0.381546       8   100   0       895s   1011      415     2     
   1.8.1  0.777375  0.311037      10   100   1      1337s   1011      415     5     
   2.2.0  0.777560  0.348724       4   100   1      4699s   4481     1065     5     
   2.1.0  0.784135  0.327790      10   100   0      8509s   4481     1065     5     
   1.9.1  0.790879  0.303636       8   100   0       895s   1011      415     5     
   2.0.0  0.793170  0.323632       4   100   0      3559s   4481     1065     5     
   3.1.0  0.797092  0.275271      10   100   0        45s    866     1065     5     
     1.9  0.801539  0.251177       8   100   0       895s   1011      415    10     
     1.6  0.802802  0.246046       5   100   0       556s   1011      415    10     
     1.7  0.803727  0.270047       5   100   1       687s   1011      415    10     
     1.5  0.804970  0.276699       6   100   1       882s   1011      415    10     
   3.0.0  0.805366  0.261353       4   100   0        19s    866     1065     5     
     1.8  0.808020  0.267684      10   100   1      1337s   1011      415    10     
     1.4  0.811572  0.245559       6   100   0       687s   1011      415    10     
     1.2  0.817391  0.274402       3   100   1       435s   1011      415    10     
     1.3  0.817779  0.232192       3   100   0       347s   1011      415    10     
     1.1  0.828783  0.233289       1   100   0       118s   1011      415    10     
     1.0  0.833424  0.258153       1   100   1       151s   1011      415    10     
    1.10  0.845832  0.215527       1    10   0       112s   1011      415    10 

Versions

<dataset>.<iteration>.<variation>

Dataset 1

Labeled and unlabled terms from the MoMA dataset.

Dataset 2

Labeled and unlabled terms from the MoMA and DOME datasets.

Dataset 3

Only labled terms from the MoMA and DOME datasets.

Usage

# Create or load an existing version.
# Versions follow the format `<dataset>.<iteration>.<variation>`.
>>> pipe = Pipeline('9.0.0')

# The current state
>>> pipe.get_state()
<class 'import_exhibitions.ImportExhibitions'>

# Get the version configurations
>>> pipe.version.config
{'combinations_r': 5,
 'epochs': 5,
 'min_count': 1,
 'pos': 0,
 'sg': 1,
 'size': 100,
 'states': ['ImportExhibitions',
            'Prune1',
            'LabelArtists',
            'Prune2',
            'ExportCorpus',
            'TrainModel',
            'ApplySimilar',
            'Report'],
 'topn': 5,
 'train_dir': 'data/train-9',
 'version': '9.0.0',
 'version_dir': 'data/versions/9.0.0',
 'workers': 5}

# Update the version configurations
>>> pipe.version.update_config(
  ('workers', 6), 
  ('size', 200), 
  ('sg', 0))

# Execute the current state and proceed to the next.
>>> pipe.proceed()