def plot_type_evaluation():
"""
Plot results from the :func:`evaluate_dep_types` experiment.
"""
l = [
'dep', 'aux', 'auxpass', 'cop', 'agent', 'acomp', 'attr', 'ccomp',
'xcomp', 'complm', 'dobj', 'iobj', 'pobj', 'mark', 'rel', 'nsubj',
'nsubjpass', 'csubj', 'csubjpass', 'cc', 'conj', 'expl', 'abbrev',
'amod', 'appos', 'advcl', 'purpcl', 'det', 'predet', 'preconj',
'infmod', 'mwe', 'partmod', 'advmod', 'neg', 'rcmod', 'quantmod',
'tmod', 'nn', 'npadvmod', 'num', 'number', 'prep', 'poss',
'possessive', 'prt', 'parataxis', 'punct', 'ref', 'xsubj', 'pcomp',
'prepc'
]
d = data.pickle_from_file('output/dependencies/types_eval_class')
diffs = []
print '--- Classification ---'
for i, dep_type in enumerate(l):
val = d['classification'][i]
diff = val - 0.5750
diffs.append(diff)
print "\t\t\t" + dep_type + " & " + '%1.4f' % val + " & " + '%1.4f' % diff + "\\\\"
d = data.pickle_from_file('output/dependencies/types_eval_retr')
diffs = []
print '--- Retrieval ---'
for i, dep_type in enumerate(l):
val = d['retrieval'][i]
diff = val - 0.1985
diffs.append(diff)
print "\t\t\t" + dep_type + " & " + '%1.4f' % val + " & " + '%1.4f' % diff + "\\\\"
def plot_type_evaluation():
"""
Plot results from the :func:`evaluate_dep_types` experiment.
"""
l = ['dep', 'aux', 'auxpass', 'cop', 'agent', 'acomp',
'attr', 'ccomp', 'xcomp', 'complm', 'dobj', 'iobj',
'pobj', 'mark', 'rel', 'nsubj', 'nsubjpass', 'csubj',
'csubjpass', 'cc', 'conj', 'expl', 'abbrev', 'amod',
'appos', 'advcl', 'purpcl', 'det', 'predet', 'preconj',
'infmod', 'mwe', 'partmod', 'advmod', 'neg', 'rcmod',
'quantmod', 'tmod', 'nn', 'npadvmod', 'num', 'number',
'prep', 'poss', 'possessive', 'prt', 'parataxis',
'punct', 'ref', 'xsubj', 'pcomp', 'prepc']
d = data.pickle_from_file('output/dependencies/types_eval_class')
diffs = []
print '--- Classification ---'
for i, dep_type in enumerate(l):
val = d['classification'][i]
diff = val - 0.5750
diffs.append(diff)
print "\t\t\t"+dep_type+" & "+'%1.4f'%val+" & "+'%1.4f'%diff +"\\\\"
d = data.pickle_from_file('output/dependencies/types_eval_retr')
diffs = []
print '--- Retrieval ---'
for i, dep_type in enumerate(l):
val = d['retrieval'][i]
diff = val - 0.1985
diffs.append(diff)
print "\t\t\t"+dep_type+" & "+'%1.4f'%val+" & "+'%1.4f'%diff +"\\\\"
def print_hubs():
"""
Print results from :func:`print_common_hub_words` as latex table.
"""
w = data.pickle_from_file('output/dependencies/common_hubs_withstop_words')
wo = data.pickle_from_file('output/dependencies/common_hubs_withoutstop_words')
tasa_w = [term.encode('ascii','ignore') for term in w['tasa'][:10]]
air_w = [term.encode('ascii','ignore') for term in w['air'][:10]]
tasa_wo = [term.encode('ascii','ignore') for term in wo['tasa'][:10]]
air_wo = [term.encode('ascii','ignore') for term in wo['air'][:10]]
for i in range(10):
print tasa_w[i],' & ',air_w[i],' & ',tasa_wo[i],' & ',air_wo[i],'\\\\'
def plot_results():
retr_results = data.pickle_from_file('output/retr_context_10')
retr_results = {
'Degree (window)': [
0.22290305491606582, 0.2239404496699994, 0.22351183191703122,
0.22293583927185456, 0.2216027852882311, 0.22232860216650002,
0.22230162622918934, 0.22287683186704185, 0.22266252053221772,
0.22237418794670616
],
'PageRank (window)': [
0.21772129149181993, 0.21884861149427587, 0.22063142971295358,
0.21893898241891538, 0.21973766615441442, 0.22054672890564322,
0.22099589130745473, 0.22129686184085004, 0.22148942934157456,
0.22147928890310792
],
'PageRank (sentence)': [0.22056586008664569] * 10,
'Degree (sentence)': [0.21784622825075944] * 10
}
#~ #'PageRank (sentence)':[0.223649757653]*10,
#~ #'Weighted degree (sentence)':[0.223449136101]*10}
pp.pprint(retr_results)
plotter.plot(range(1, 11),
retr_results,
'retrieval score',
'n, context size',
'', [1, 10, .216, .225],
legend_place="lower right")
def test_co_occurrences():
doc1 = data.read_file('../data/tasa/TASATest/Science/Agatha09.07.03.txt')
doc2 = data.read_file('../data/tasa/TASATest_preprocessed/Science/Agatha09.07.03.txt')
g0 = construct_cooccurrence_network(doc1, context='window', already_preprocessed=False)
g1 = construct_cooccurrence_network(doc2, context='window', already_preprocessed=True)
g2 = construct_cooccurrence_network(doc1, context='sentence', already_preprocessed=False)
graphs = data.pickle_from_file('output/testdata/co-occurrence-graphs.pkl')
assert(graph.equal(g0,graphs[0]))
assert(graph.equal(g1,graphs[1]))
assert(graph.equal(g2,graphs[2]))
doc = data.read_file('output/testdata/higher.order.testdoc.preprocessed.txt')
g1 = construct_cooccurrence_network(doc, already_preprocessed=True, window_size=1, orders=[1])
g12 = construct_cooccurrence_network(doc, already_preprocessed=True, window_size=1, orders=[1,2])
g123 = construct_cooccurrence_network(doc, already_preprocessed=True, window_size=1, orders=[1,2,3])
g13 = construct_cooccurrence_network(doc, already_preprocessed=True, window_size=1, orders=[1,3])
assert(('foo','bar') in g1.edges())
assert(('foo','baz') not in g1.edges())
assert(('foo','cake') not in g1.edges())
assert(('foo','bar') in g12.edges())
assert(('foo','baz') in g12.edges())
assert(('foo','cake') not in g12.edges())
assert(('foo','bar') in g123.edges())
assert(('foo','baz') in g123.edges())
assert(('foo','cake') in g123.edges())
assert(('foo','baz') not in g13.edges())
print 'ok'
def print_hubs():
"""
Print results from :func:`print_common_hub_words` as latex table.
"""
w = data.pickle_from_file('output/dependencies/common_hubs_withstop_words')
wo = data.pickle_from_file(
'output/dependencies/common_hubs_withoutstop_words')
tasa_w = [term.encode('ascii', 'ignore') for term in w['tasa'][:10]]
air_w = [term.encode('ascii', 'ignore') for term in w['air'][:10]]
tasa_wo = [term.encode('ascii', 'ignore') for term in wo['tasa'][:10]]
air_wo = [term.encode('ascii', 'ignore') for term in wo['air'][:10]]
for i in range(10):
print tasa_w[i], ' & ', air_w[i], ' & ', tasa_wo[i], ' & ', air_wo[
i], '\\\\'
def plot_results():
retr_results = data.pickle_from_file('output/retr_context_10')
retr_results = {'Degree (window)': [0.22290305491606582,
0.2239404496699994,
0.22351183191703122,
0.22293583927185456,
0.2216027852882311,
0.22232860216650002,
0.22230162622918934,
0.22287683186704185,
0.22266252053221772,
0.22237418794670616],
'PageRank (window)': [0.21772129149181993,
0.21884861149427587,
0.22063142971295358,
0.21893898241891538,
0.21973766615441442,
0.22054672890564322,
0.22099589130745473,
0.22129686184085004,
0.22148942934157456,
0.22147928890310792],
'PageRank (sentence)': [0.22056586008664569]*10,
'Degree (sentence)': [0.21784622825075944]*10}
#~ #'PageRank (sentence)':[0.223649757653]*10,
#~ #'Weighted degree (sentence)':[0.223449136101]*10}
pp.pprint(retr_results)
plotter.plot(range(1,11),retr_results,'retrieval score','n, context size','',[1,10,.216,.225], legend_place="lower right")
def retrieval_comparison_graph(dataset='air', graph_type='co-occurrence', use_icc=False):
"""
Experiment used for comparative evaluation of different network
representations on retrieval.
graph_type = 'co-occurrence' | 'dependency'
`icc` determines whether to use _inverse corpus centrality_ in the vector representations.
"""
def make_dicts(docs, icc=None):
rep = []
for i, doc in enumerate(docs):
if i%100==0: print ' graph',str(i)+'/'+str(len(docs))
g = gfuns[graph_type](doc)
d = graph_representation.graph_to_dict(g, metrics[graph_type], icc)
rep.append(d)
return rep
postfix = {'co-occurrence':'_text', 'dependency':'_dependencies'}
gfuns = {'co-occurrence':graph_representation.construct_cooccurrence_network,
'dependency':graph_representation.construct_dependency_network}
metrics = {'co-occurrence':graph.GraphMetrics.WEIGHTED_DEGREE,
'dependency':graph.GraphMetrics.EIGENVECTOR}
print '--', graph_type
print '> Reading data..', dataset
path = '../data/'+dataset+'/problem_descriptions'+postfix[graph_type]
docs, labels = data.read_files(path)
print '> Creating solution representations..'
solutions_path = '../data/'+dataset+'/solutions_preprocessed'
solutions_texts, labels = data.read_files(solutions_path)
solutions_rep = freq_representation.text_to_vector(solutions_texts, freq_representation.FrequencyMetrics.TF_IDF)
icc = None
if use_icc:
print '> Calculating ICC..'
m = metrics[graph_type].split()[0]
print graph_type
if graph_type == 'co-occurrence':
p = 'output/centralities/co-occurrence/'+dataset+'/problem_descriptions/window/'+m+'.cent'
elif graph_type == 'dependency':
p = 'output/centralities/dependency/'+dataset+'/problem_descriptions/'+m+'.cent'
print ' fetching', p
icc = data.pickle_from_file(p)
print ' icc:', type(icc)
print '> Creating problem description representations..'
dicts = make_dicts(docs, icc)
descriptions_rep = graph_representation.dicts_to_vectors(dicts)#, remove_stop_words=True)
print '> Evaluating..'
results = evaluation.evaluate_retrieval(descriptions_rep, solutions_rep)
print results
s = 'retrieval comparison '
if use_icc: s += 'USING TC-ICC'
s += '\nrepresentation: '+graph_type+'\nresult: '+str(results)+'\n\n\n'
data.write_to_file(s, 'output/comparison/retrieval')
return results
def compare_stats_to_random(dataset):
dataset = dataset.replace('/','.')
stats = data.pickle_from_file('output/properties/cooccurrence/stats_tot_'+dataset)
n = stats['# nodes_mean']
p = stats['mean degree_mean']/(2*n)
g = nx.directed_gnp_random_graph(int(n), p)
props = graph.network_properties(g)
pp.pprint(props)
def compare_stats_to_random(dataset):
dataset = dataset.replace('/', '.')
stats = data.pickle_from_file('output/properties/cooccurrence/stats_tot_' +
dataset)
n = stats['# nodes_mean']
p = stats['mean degree_mean'] / (2 * n)
g = nx.directed_gnp_random_graph(int(n), p)
props = graph.network_properties(g)
pp.pprint(props)
def construct_cooccurrence_network(doc,
window_size=2,
direction='undirected',
context='sentence',
already_preprocessed=False,
orders=[],
order_weights=[1.0, 1.0, 1.0],
doc_id=None,
verbose=False):
"""Construct co-occurrence network from text.
*direction* must be 'forward', 'backward' or 'undirected', while *context*
can be 'window' or 'sentence'.
If *context* is 'window', *already_preprocessed* indicate whether *doc*
already have been processed. Sentence contexts require unpreocessed *doc*s.
Any value for *window_size* is ignored if *context* is 'sentence'.
A DiGraph is created regardless of direction parameter, but with 'undirected',
edges are created in both directions.
"""
doc = _cooccurrence_preprocessing(doc, context, already_preprocessed)
if context is 'sentence':
matrix, term_list = _sentence_cooccurrence_matrix(
doc, direction, verbose)
elif context is 'window':
matrix, term_list = _window_cooccurrence_matrix(
doc, direction, window_size, verbose)
g = nx.DiGraph()
g.add_nodes_from(term_list)
if len(orders) == 0:
graph.add_edges_from_matrix(g, matrix, term_list)
else:
if doc_id is not None and os.path.exists(doc_id):
first, second, third = data.pickle_from_file(doc_id)
else:
first, second, third = _higher_order_matrix(matrix.todense())
if doc_id is not None:
data.pickle_to_file((first, second, third), doc_id)
if 1 in orders:
graph.add_edges_from_matrix(g,
first,
term_list,
rel_weight=order_weights[0])
if 2 in orders:
graph.add_edges_from_matrix(g,
second,
term_list,
rel_weight=order_weights[1])
if 3 in orders:
graph.add_edges_from_matrix(g,
third,
term_list,
rel_weight=order_weights[2])
return g
def store_corpus_network(corpus, context):
print '> Constructing corpus network for', corpus
path = '../data/'+corpus+'_text'
store_path = 'output/giants/co-occurrence/'+corpus+'/'+context+'_graph.net'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping'
return
texts, labels = data.read_files(path)
gdoc = ' '.join(texts)
giant = graph_representation.construct_cooccurrence_network(gdoc, context=context, already_preprocessed=False, verbose=True)
print '> Serializing and saving..'
data.pickle_to_file(giant, store_path)
def store_corpus_network(corpus, context):
print '> Constructing corpus network for', corpus
path = '../data/' + corpus + '_text'
store_path = 'output/giants/co-occurrence/' + corpus + '/' + context + '_graph.net'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping'
return
texts, labels = data.read_files(path)
gdoc = ' '.join(texts)
giant = graph_representation.construct_cooccurrence_network(
gdoc, context=context, already_preprocessed=False, verbose=True)
print '> Serializing and saving..'
data.pickle_to_file(giant, store_path)
def print_network_props():
"""
Prints latex table with various properties for networks created from
texts in the datasets.
"""
print '-- Co-occurrence'
tasa = data.pickle_from_file('output/properties/cooccurrence/stats_tot_tasa.TASA900')
air = data.pickle_from_file('output/properties/cooccurrence/stats_tot_air.problem_descriptions')
for key in air.keys():
prop, sep, mod = key.partition('_')
if mod!='std':
print prop,' & ',
print '%2.3f'%tasa[prop+sep+'mean'],' & ','%2.3f'%tasa[prop+sep+'std'],' & ',
print '%2.3f'%air[prop+sep+'mean'],' & ','%2.3f'%air[prop+sep+'std'],'\\\\'
print
print '-- Dependency, all types'
air = data.pickle_from_file('output/properties/dependency/stats_tot_air.problem_descriptions')
tasa = data.pickle_from_file('output/properties/dependency/stats_tot_tasa.TASA900')
for key in air.keys():
prop, sep, mod = key.partition('_')
if mod!='std':
print prop,' & ',
print '%2.3f'%tasa[prop+sep+'mean'],' & ','%2.3f'%tasa[prop+sep+'std'],' & ',
print '%2.3f'%air[prop+sep+'mean'],' & ','%2.3f'%air[prop+sep+'std'],'\\\\'
def print_network_props():
"""
Prints latex table with various properties for networks created from
texts in the datasets.
"""
print '-- Co-occurrence'
tasa = data.pickle_from_file(
'output/properties/cooccurrence/stats_tot_tasa.TASA900')
air = data.pickle_from_file(
'output/properties/cooccurrence/stats_tot_air.problem_descriptions')
for key in air.keys():
prop, sep, mod = key.partition('_')
if mod != 'std':
print prop, ' & ',
print '%2.3f' % tasa[prop + sep +
'mean'], ' & ', '%2.3f' % tasa[prop + sep +
'std'], ' & ',
print '%2.3f' % air[prop + sep +
'mean'], ' & ', '%2.3f' % air[prop + sep +
'std'], '\\\\'
print
print '-- Dependency, all types'
air = data.pickle_from_file(
'output/properties/dependency/stats_tot_air.problem_descriptions')
tasa = data.pickle_from_file(
'output/properties/dependency/stats_tot_tasa.TASA900')
for key in air.keys():
prop, sep, mod = key.partition('_')
if mod != 'std':
print prop, ' & ',
print '%2.3f' % tasa[prop + sep +
'mean'], ' & ', '%2.3f' % tasa[prop + sep +
'std'], ' & ',
print '%2.3f' % air[prop + sep +
'mean'], ' & ', '%2.3f' % air[prop + sep +
'std'], '\\\\'
def test_co_occurrences():
doc1 = data.read_file('../data/tasa/TASATest/Science/Agatha09.07.03.txt')
doc2 = data.read_file(
'../data/tasa/TASATest_preprocessed/Science/Agatha09.07.03.txt')
g0 = construct_cooccurrence_network(doc1,
context='window',
already_preprocessed=False)
g1 = construct_cooccurrence_network(doc2,
context='window',
already_preprocessed=True)
g2 = construct_cooccurrence_network(doc1,
context='sentence',
already_preprocessed=False)
graphs = data.pickle_from_file('output/testdata/co-occurrence-graphs.pkl')
assert (graph.equal(g0, graphs[0]))
assert (graph.equal(g1, graphs[1]))
assert (graph.equal(g2, graphs[2]))
doc = data.read_file(
'output/testdata/higher.order.testdoc.preprocessed.txt')
g1 = construct_cooccurrence_network(doc,
already_preprocessed=True,
window_size=1,
orders=[1])
g12 = construct_cooccurrence_network(doc,
already_preprocessed=True,
window_size=1,
orders=[1, 2])
g123 = construct_cooccurrence_network(doc,
already_preprocessed=True,
window_size=1,
orders=[1, 2, 3])
g13 = construct_cooccurrence_network(doc,
already_preprocessed=True,
window_size=1,
orders=[1, 3])
assert (('foo', 'bar') in g1.edges())
assert (('foo', 'baz') not in g1.edges())
assert (('foo', 'cake') not in g1.edges())
assert (('foo', 'bar') in g12.edges())
assert (('foo', 'baz') in g12.edges())
assert (('foo', 'cake') not in g12.edges())
assert (('foo', 'bar') in g123.edges())
assert (('foo', 'baz') in g123.edges())
assert (('foo', 'cake') in g123.edges())
assert (('foo', 'baz') not in g13.edges())
print 'ok'
def store_corpus_network(corpus):
print '> Constructing corpus network for', corpus
path = '../data/'+corpus+'_dependencies'
store_path = 'output/giants/dependency/'+corpus+'/graph.net'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping'
return
texts, labels = data.read_files(path)
gdeps = {}
for i, text in enumerate(texts):
if i%1==0: print ' ',str(i)+'/'+str(len(texts))
d = pickle.loads(text)
for dep in d.keys():
gdeps[dep] = gdeps.get(dep, []) + d[dep]
giant = graph_representation.construct_dependency_network(gdeps,verbose=True,unpickle=False)
print '> Serializing and saving..'
data.pickle_to_file(giant, store_path)
def store_corpus_network(corpus):
print '> Constructing corpus network for', corpus
path = '../data/' + corpus + '_dependencies'
store_path = 'output/giants/dependency/' + corpus + '/graph.net'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping'
return
texts, labels = data.read_files(path)
gdeps = {}
for i, text in enumerate(texts):
if i % 1 == 0: print ' ', str(i) + '/' + str(len(texts))
d = pickle.loads(text)
for dep in d.keys():
gdeps[dep] = gdeps.get(dep, []) + d[dep]
giant = graph_representation.construct_dependency_network(gdeps,
verbose=True,
unpickle=False)
print '> Serializing and saving..'
data.pickle_to_file(giant, store_path)
def store_centralities(corpus, context):
print '> Calculating and storing centralities for', corpus
g = retrieve_corpus_network(corpus, context)
metrics = graph_representation.get_metrics(True, exclude_flow=True)
for metric in metrics:
m = metric.split()[0]
store_path = 'output/centralities/co-occurrence/'+corpus+'/'+context+'/'+m+'.cent'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping:', metric
continue
else:
print ' calculating:', metric
try:
c = graph.centralities(g, metric)
data.pickle_to_file(c, store_path)
except MemoryError as e:
print 'MemoryError :('
data.write_to_file('MemoryError while claculating '+metric+' on '+corpus+':\n'+str(e)+'\n\n', 'output/log/errors')
def plot_sentence_lengths(datafile=None):
"""
Function for plotting histogram of sentence lengths within a given dataset.
"""
if datafile is None:
import preprocess
print '> reading data..'
path = '../data/tasa/TASA900_text'
texts, labels = data.read_files(path)
sentence_lengths = []
print '> counting lengths..'
for text in texts:
sentences = preprocess.tokenize_sentences(text)
for sentence in sentences:
tokens = preprocess.tokenize_tokens(sentence)
sentence_lengths.append(len(tokens))
data.pickle_to_file(sentence_lengths, 'output/tasa_sentence_lengths.pkl')
else:
sentence_lengths = data.pickle_from_file(datafile)
plotter.histogram(sentence_lengths, 'sentence length (tokens)', '# sentences', bins=70)
def store_centralities(corpus, context):
print '> Calculating and storing centralities for', corpus
g = retrieve_corpus_network(corpus, context)
metrics = graph_representation.get_metrics(True, exclude_flow=True)
for metric in metrics:
m = metric.split()[0]
store_path = 'output/centralities/co-occurrence/' + corpus + '/' + context + '/' + m + '.cent'
if data.pickle_from_file(store_path, suppress_warning=True):
print ' already present, skipping:', metric
continue
else:
print ' calculating:', metric
try:
c = graph.centralities(g, metric)
data.pickle_to_file(c, store_path)
except MemoryError as e:
print 'MemoryError :('
data.write_to_file(
'MemoryError while claculating ' + metric + ' on ' + corpus +
':\n' + str(e) + '\n\n', 'output/log/errors')
def construct_cooccurrence_network(doc, window_size=2, direction='undirected', context='sentence', already_preprocessed=False, orders=[], order_weights=[1.0,1.0,1.0],doc_id=None,verbose=False):
"""Construct co-occurrence network from text.
*direction* must be 'forward', 'backward' or 'undirected', while *context*
can be 'window' or 'sentence'.
If *context* is 'window', *already_preprocessed* indicate whether *doc*
already have been processed. Sentence contexts require unpreocessed *doc*s.
Any value for *window_size* is ignored if *context* is 'sentence'.
A DiGraph is created regardless of direction parameter, but with 'undirected',
edges are created in both directions.
"""
doc = _cooccurrence_preprocessing(doc, context, already_preprocessed)
if context is 'sentence':
matrix, term_list = _sentence_cooccurrence_matrix(doc, direction, verbose)
elif context is 'window':
matrix, term_list = _window_cooccurrence_matrix(doc, direction, window_size, verbose)
g = nx.DiGraph()
g.add_nodes_from(term_list)
if len(orders)==0:
graph.add_edges_from_matrix(g, matrix, term_list)
else:
if doc_id is not None and os.path.exists(doc_id):
first, second, third = data.pickle_from_file(doc_id)
else:
first, second, third = _higher_order_matrix(matrix.todense())
if doc_id is not None:
data.pickle_to_file((first,second,third), doc_id)
if 1 in orders:
graph.add_edges_from_matrix(g, first, term_list, rel_weight=order_weights[0])
if 2 in orders:
graph.add_edges_from_matrix(g, second, term_list, rel_weight=order_weights[1])
if 3 in orders:
graph.add_edges_from_matrix(g, third, term_list, rel_weight=order_weights[2])
return g
def plot_sentence_lengths(datafile=None):
"""
Function for plotting histogram of sentence lengths within a given dataset.
"""
if datafile is None:
import preprocess
print '> reading data..'
path = '../data/tasa/TASA900_text'
texts, labels = data.read_files(path)
sentence_lengths = []
print '> counting lengths..'
for text in texts:
sentences = preprocess.tokenize_sentences(text)
for sentence in sentences:
tokens = preprocess.tokenize_tokens(sentence)
sentence_lengths.append(len(tokens))
data.pickle_to_file(sentence_lengths,
'output/tasa_sentence_lengths.pkl')
else:
sentence_lengths = data.pickle_from_file(datafile)
plotter.histogram(sentence_lengths,
'sentence length (tokens)',
'# sentences',
bins=70)
def retrieve_centralities(corpus, context, metric):
m = metric.split()[0]
path = 'output/centralities/co-occurrence/' + corpus + '/' + context + '/' + m + '.cent'
print ' retrieving', path
return data.pickle_from_file(path)
def retrieve_corpus_network(corpus, context):
path = 'output/giants/co-occurrence/' + corpus + '/' + context + '_graph.net'
return data.pickle_from_file(path)
def retrieve_corpus_network(corpus):
path = 'output/giants/dependency/' + corpus + '/graph.net'
return data.pickle_from_file(path)
def retrieve_corpus_network(corpus):
path = 'output/giants/dependency/'+corpus+'/graph.net'
return data.pickle_from_file(path)
def retrieval_comparison_graph(dataset='air',
graph_type='co-occurrence',
use_icc=False):
"""
Experiment used for comparative evaluation of different network
representations on retrieval.
graph_type = 'co-occurrence' | 'dependency'
`icc` determines whether to use _inverse corpus centrality_ in the vector representations.
"""
def make_dicts(docs, icc=None):
rep = []
for i, doc in enumerate(docs):
if i % 100 == 0: print ' graph', str(i) + '/' + str(len(docs))
g = gfuns[graph_type](doc)
d = graph_representation.graph_to_dict(g, metrics[graph_type], icc)
rep.append(d)
return rep
postfix = {'co-occurrence': '_text', 'dependency': '_dependencies'}
gfuns = {
'co-occurrence': graph_representation.construct_cooccurrence_network,
'dependency': graph_representation.construct_dependency_network
}
metrics = {
'co-occurrence': graph.GraphMetrics.WEIGHTED_DEGREE,
'dependency': graph.GraphMetrics.EIGENVECTOR
}
print '--', graph_type
print '> Reading data..', dataset
path = '../data/' + dataset + '/problem_descriptions' + postfix[graph_type]
docs, labels = data.read_files(path)
print '> Creating solution representations..'
solutions_path = '../data/' + dataset + '/solutions_preprocessed'
solutions_texts, labels = data.read_files(solutions_path)
solutions_rep = freq_representation.text_to_vector(
solutions_texts, freq_representation.FrequencyMetrics.TF_IDF)
icc = None
if use_icc:
print '> Calculating ICC..'
m = metrics[graph_type].split()[0]
print graph_type
if graph_type == 'co-occurrence':
p = 'output/centralities/co-occurrence/' + dataset + '/problem_descriptions/window/' + m + '.cent'
elif graph_type == 'dependency':
p = 'output/centralities/dependency/' + dataset + '/problem_descriptions/' + m + '.cent'
print ' fetching', p
icc = data.pickle_from_file(p)
print ' icc:', type(icc)
print '> Creating problem description representations..'
dicts = make_dicts(docs, icc)
descriptions_rep = graph_representation.dicts_to_vectors(
dicts) #, remove_stop_words=True)
print '> Evaluating..'
results = evaluation.evaluate_retrieval(descriptions_rep, solutions_rep)
print results
s = 'retrieval comparison '
if use_icc: s += 'USING TC-ICC'
s += '\nrepresentation: ' + graph_type + '\nresult: ' + str(
results) + '\n\n\n'
data.write_to_file(s, 'output/comparison/retrieval')
return results
def retrieve_centralities(corpus, context, metric):
m = metric.split()[0]
path = 'output/centralities/co-occurrence/'+corpus+'/'+context+'/'+m+'.cent'
print ' retrieving',path
return data.pickle_from_file(path)
def retrieve_corpus_network(corpus, context):
path = 'output/giants/co-occurrence/'+corpus+'/'+context+'_graph.net'
return data.pickle_from_file(path)
def retrieve_centralities(corpus, metric):
m = metric.split()[0]
path = 'output/centralities/dependency/' + corpus + '/' + m + '.cent'
print ' retrieving', path
return data.pickle_from_file(path)
def retrieve_centralities(corpus, metric):
m = metric.split()[0]
path = 'output/centralities/dependency/'+corpus+'/'+m+'.cent'
print ' retrieving',path
return data.pickle_from_file(path)