2018-11-08¶Agglomerative clustering, test_grammar updated 2018-10-19 , Link Grammar 5.4.4.
This notebook is shared as static Child-Directed-Speech-ALE-2018-11-08_.html
Data -- Child-Directed-Speech-ALE-2018-11-08_ directory.
Previous (reference) tests:
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
from src.grammar_learner.read_files import check_dir
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
tmpath = module_path + '/tmp/'
check_dir(tmpath, True, 'none')
table = []
long_table = []
start = time.time()
print(UTC(), ':: module_path =', module_path)
corpus = 'CDS-caps-br-text+brent9mos'
corpus = 'CDS-caps-br-text'
dataset = 'LG-English'
dataset = 'LG-English-clean-clean' # 2018-10-29: only 100% parsed
out_dir = module_path + '/output/Child-Directed-Speech-ALE-' + str(UTC())[:10] + '_'
runs = (1,1)
if runs != (1,1): out_dir += '-multi'
kwargs = {
'left_wall' : '' ,
'period' : False ,
'context' : 2 ,
'min_word_count': 1 ,
'min_link_count': 1 ,
'max_words' : 100000 ,
'max_features' : 100000 ,
'min_co-occurrence_count': 1 ,
'min_co-occurrence_probability': 1e-9,
'word_space' : 'sparse' ,
'clustering' : ('agglomerative', 'ward'),
'cluster_range' : (20,200,20,1),
'cluster_criteria' : 'silhouette',
'clustering_metric' : ('silhouette', 'cosine'),
'cluster_level' : 1 ,
'grammar_rules' : 2 ,
'max_disjuncts' : 100000 ,
'tmpath' : tmpath ,
'verbose' : 'min' ,
'template_path' : 'poc-turtle',
'linkage_limit' : 1000 ,
'categories_generalization': 'off' }
lines = [
[33, corpus , 'LG-English' ,0,0, 'none' ],
[34, corpus , 'LG-English' ,0,0, 'rules' ],
[35, corpus , 'R=6-Weight=6:R-mst-weight=+1:R' ,0,0, 'none' ],
[36, corpus , 'R=6-Weight=6:R-mst-weight=+1:R' ,0,0, 'rules' ]]
# rp = module_path + '/data/CDS-caps-br-text+brent9mos/LG-English'
rp = module_path + '/data/CDS-caps-br-text/LG-English' # shorter test 81025
rp = module_path + '/data/CDS-caps-br-text/LG-English-clean-clean'
cp = rp # corpus path = reference_path :: use 'gold' parses as test corpus
%%capture
kwargs['cluster_range'] = 200
line = [lines[0]]
linez = [lines[0], lines[1]]
out = out_dir + '/200-clusters'
a, _, header = table_rows(lines, out, cp, rp, runs, **kwargs)
display(html_table([header] + a))
display(html_table([header] + a))
"clean-clean" dataset -- all incomplete parses removed
%%capture
corpus = 'CDS-caps-br-text'
dataset = 'LG-English-clean-clean' # 2018-10-29: only 100% parsed sentences
out = out_dir + '/clean-training-set'
t1 = []
table = []
crange = kwargs['cluster_range']
for kwargs['cluster_range'] in range(20,201,20):
average, _, header = table_rows(line, out, cp, rp, runs, **kwargs)
t1.extend(average)
table.extend(average)
kwargs['cluster_range'] = crange
display(html_table([header] + t1))
%%capture
corpus = 'CDS-caps-br-text'
dataset = 'LG-English' # initial parses, no cleanup
out = out_dir + '/basic-training-set'
t2 = []
for kwargs['cluster_range'] in range(20,201,20):
average, _, header = table_rows(line, out, cp, rp, runs, **kwargs)
t2.extend(average)
table.extend(average)
display(html_table([header] + t2))