Mean shift clustering "Gutenberg Children Books" `2019-03-28`¶

"Gutenberg Children Books" corpus, new "LG-E-noQuotes" dataset (GC_LGEnglish_noQuotes_fullyParsed.ull),
trash filter off: min_word_count = 31,21,11,6,2,1, max_sentence_length off, Link Grammar 5.6.

This notebook is shared as static Mean-shift-clustering-GCB-LG-E-noQuotes-2019-03-28.html.
Output data shared via Mean-shift-clustering-GCB-LG-E-noQuotes-2019-03-28 directory.

Basic settings¶

import os, sys, time
module_path = os.path.abspath(os.path.join('..'))
if module_path not in sys.path: sys.path.append(module_path)
from src.grammar_learner.utl import UTC, test_stats
from src.grammar_learner.read_files import check_dir, check_corpus
from src.grammar_learner.write_files import list2file
from src.grammar_learner.widgets import html_table
from src.grammar_learner.pqa_table import table_rows, params, wide_rows
tmpath = module_path + '/tmp/'
check_dir(tmpath, True, 'none')
start = time.time()
runs = (1,1)
print(UTC(), ':: module_path:', module_path)

2019-03-28 06:36:08 UTC :: module_path: /home/obaskov/94/ULL

Corpus test settings¶

corpus = 'GCB' # 'Gutenberg-Children-Books-Caps' 
dataset = 'LG-E-noQuotes'  # 'LG-E-clean'
kwargs = {
    'left_wall'     :   ''          ,
    'period'        :   False       ,
    'context'       :   1           ,
    'min_word_count':   1           ,  # 11, 1
    'word_space'    :   'sparse'    ,
    'clustering'    :   ('mean_shift', 2),
    'clustering_metric' : ['silhouette', 'cosine'],
    'cluster_range' :   [0]         ,   # auto
    'top_level'     :   0.01        ,
    'grammar_rules' :   2           ,
    'max_disjuncts' :   1000000     ,   # off
    'stop_words'    :   []          ,
    'tmpath'        :   tmpath      ,
    'verbose'       :   'log+'      ,
    'template_path' :   'poc-turtle',
    'linkage_limit' :   1000        }
rp = module_path + '/data/GCB/LG-E-noQuotes/'
cp = rp  # corpus path = reference_path
out_dir = module_path + '/output/' + 'Mean-shift-GCB-LG-noQuotes-' + str(UTC())[:10]
print(UTC(), '\n', out_dir)

2019-03-28 06:36:08 UTC 
 /home/obaskov/94/ULL/output/Mean-shift-GCB-LG-noQuotes-2019-03-28

Tests: `min_word_count = 31, 21, 11, 6, 2`¶

%%capture
table = []
line = [['', corpus, dataset, 0, 0, 'none']]
kwargs['min_word_count'] = 31
a, _, header, log, rules = wide_rows(line, out_dir, cp, rp, runs, **kwargs)
header[0] = ''
table.extend(a)

display(html_table([header] + a)); print(test_stats(log))

Cleaned dictionary: 3341 words, grammar learn time: 00:34:11, grammar test time: 00:18:14

display(html_table([header] + a)); print(test_stats(log))

Cleaned dictionary: 4417 words, grammar learn time: 01:03:23, grammar test time: 00:19:23

display(html_table([header] + a)); print(test_stats(log))

Cleaned dictionary: 6866 words, grammar learn time: 03:05:49, grammar test time: 00:20:23

display(html_table([header] + a)); print(test_stats(log))

Cleaned dictionary: 10053 words, grammar learn time: 09:18:28, grammar test time: 00:22:33

display(html_table([header] + a)); print(test_stats(log))

Cleaned dictionary: 19326 words, grammar learn time: 71:29:28, grammar test time: 00:21:04

Save results¶

display(html_table([header] + table))

print(UTC(), ':: 5 tests finished, elapsed', str(round((time.time()-start)/3600.0, 1)), 'hours')
table_str = list2file(table, out_dir + '/all_tests_table.txt')
print('Results saved to', out_dir + '/all_tests_table.txt')

2019-03-31 21:49:11 UTC :: 5 tests finished, elapsed 87.2 hours
Results saved to /home/obaskov/94/ULL/output/Mean-shift-GCB-LG-noQuotes-2019-03-28/all_tests_table.txt

Test with `min_word_count = 1`: 110 hours, F1 ~0.66¶

%%capture
kwargs['min_word_count'] = 1
a, _, h, log, rules = wide_rows(line, out_dir, cp, rp, runs, **kwargs)
table.extend(a)

display(html_table([header] + a)); print(test_stats(log))

Test finished 2019-04-05 10:50 UTC:
Clean corpus size 22641, Grammar learn time 110:27:19, Grammar test time 00:27:12.
PA = 63%, PQ = 60%, F1 = 0.66, Top 5 cluster sizes: [14527, 4, 3, 2, 1]

Corpus	Parsing	Space	Linkage	Affinity	G12n	Threshold	Rules	MWC	NN	SI	PA	PQ	F1	Top 5 cluster sizes
GCB	LG-E-noQuotes	cMLEd	mean_shift	---	none	---	3341	31	---	0.0	57%	52%	0.61	[1, 0]
GCB	LG-E-noQuotes	cMLEd	mean_shift	---	none	---	4417	21	---	0.0	59%	54%	0.63	[1, 0]
GCB	LG-E-noQuotes	cMLEd	mean_shift	---	none	---	6801	11	---	0.0	62%	57%	0.65	[57, 3, 2, 1, 0]
GCB	LG-E-noQuotes	cMLEd	mean_shift	---	none	---	8735	6	---	0.0	63%	59%	0.67	[1251, 7, 4, 3, 2]
GCB	LG-E-noQuotes	cMLEd	mean_shift	---	none	---	8089	2	---	0.0	64%	61%	0.67	[11194, 4, 3, 2, 1]

Mean shift clustering "Gutenberg Children Books" 2019-03-28¶