def solvegraph_leave_one_lecture_out(phrasedir, modelname='svr'):
lectures = annotation.Lectures
oslom = OSLOM(oslom_parms)
if modelname == 'svm':
weighted = False
undirect = True
elif modelname == 'lsa':
weighted = False
undirect = True
else: #svr, lsa
weighted = True
undirect = True
for i, lec in enumerate(lectures):
path = os.path.join(phrasedir, str(lec))
for q in ['q1', 'q2']:
#write the output
phrasefile = os.path.join(path, "%s.%s.key" % (q, method))
phrases = fio.LoadList(phrasefile)
netgraphfile = os.path.join(
path, "%s.%s.%s%s" % (q, method, modelname, net_exe))
oslom.solve_graph(netgraphfile, undirect, weighted)
def extractPhrasePaireFeature(phrasedir):
for lec in annotation.Lectures:
path = phrasedir + str(lec) + '/'
fio.NewPath(path)
for prompt in ['q1', 'q2']:
prefix = os.path.join(path, '%s.%s.' % (prompt, method))
filename = path + prompt + sim_exe
print filename
featureset = []
feature_extractor = Similarity(prefix)
phrasefile = os.path.join(path, "%s.%s.key" % (prompt, method))
phrases = fio.LoadList(phrasefile)
for p1 in phrases:
for p2 in phrases:
featureset.append(
(feature_extractor.get_features(p1, p2), 0.0, {
'p1': p1,
'p2': p2
}))
fio.SaveDict2Json(featureset, filename)
feature_extractor.save()
def __init__(self, key_prefix, sum_prefix, N):
'''
N is number of annotators
'''
self.key_prefix = key_prefix
self.sum_prefix = sum_prefix
self.N = N
#load phrase color map
phrasefile = key_prefix + phrase_exe
phrases = fio.LoadList(phrasefile)
colorfile = key_prefix + color_exe
color_map = fio.LoadDictJson(colorfile)
phrase_color_map = self.combine_phrase_color(phrases, color_map)
#get phrase summary color map
sumfile = sum_prefix + sum_exe
summaries = fio.LoadList(sumfile)
self.summary_color = self.get_summary_color(summaries, phrase_color_map)
#get summary count
sumcountfile = sum_prefix + sum_count_exe
self.summary_no = [int(x) for x in fio.LoadList(sumcountfile)]
assert(len(self.summary_color) == len(self.summary_no))
#load human_summary color map
self.ref_color = []
for i in range(N):
d = {}
ref_sumcolor_file = '%s%s.%d.color'%(sum_prefix, ref_exe, i)
ref_sumno_file = '%s%s.%d.no'%(sum_prefix, ref_exe, i)
for color, no in zip(fio.LoadList(ref_sumcolor_file), fio.LoadList(ref_sumno_file)):
d[int(color)] = int(no)
self.ref_color.append(d)
def writegraph_leave_one_lecture_out_lsa(model_dir,
phrasedir,
modelname='lsa'):
lectures = annotation.Lectures
for i, lec in enumerate(lectures):
test = [lec]
path = os.path.join(phrasedir, str(lec))
for q in ['q1', 'q2']:
#write the output
phrasefile = os.path.join(path, "%s.%s.key" % (q, method))
phrases = fio.LoadList(phrasefile)
if modelname == 'lsa':
similarties_results = os.path.join(
path, "%s.%s.optimumComparerLSATasa" % (q, method))
elif modelname == 'svm':
similarties_results = os.path.join(path,
"%s.%s.svm" % (q, method))
simhead, simbody = fio.ReadMatrix(similarties_results,
hasHead=True)
assert (len(simhead) == len(phrases))
body = []
for i, p1 in enumerate(phrases):
for j, p2 in enumerate(phrases):
if j <= i:
continue #undirect graph
score = simbody[i][j]
score = float(score) if score != 'NaN' else 0.0
#if score == 0.0: score = 0.000001
#if score < 0.5: continue
if score == 0.0: continue
#row = [i, j, '%f'%score]
row = [i, j]
body.append(row)
output = os.path.join(
path, "%s.%s.%s%s" % (q, method, modelname, net_exe))
fio.WriteMatrix(output, body)
def predict_IE256(train_course, model_dir, phrasedir, modelname='svm'):
sim_extractor = Similarity()
allfeatures = sorted(sim_extractor.features.keys())
features = allfeatures
name = '_'.join(features)
lectures = annotation.Lectures
for i, lec in enumerate(lectures):
test = [lec]
print test
model_file = os.path.join(model_dir,
'%s_%s.model' % (train_course, name))
with open(model_file, 'rb') as handle:
clf = pickle.load(handle)
path = os.path.join(phrasedir, str(lec))
for q in ['q1', 'q2']:
test_X, test_Y = combine_files_test(phrasedir,
test,
features,
prompts=[q])
predict_Y = clf.predict(test_X)
#write the output
phrasefile = os.path.join(path, "%s.%s.key" % (q, method))
phrases = fio.LoadList(phrasefile)
assert (len(predict_Y) == len(phrases) * len(phrases))
k = 0
body = []
for p1 in phrases:
row = []
for p2 in phrases:
row.append(predict_Y[k])
k += 1
body.append(row)
output = os.path.join(path, "%s.%s.%s" % (q, method, modelname))
fio.WriteMatrix(output, body, phrases)
def readgraph_leave_one_lecture_out(phrasedir, modelname='svr'):
lectures = annotation.Lectures
oslom = OSLOM()
if modelname == 'svr':
weighted = True
undirect = True
else:
weighted = False
undirect = True
for i, lec in enumerate(lectures):
path = os.path.join(phrasedir, str(lec))
for q in ['q1', 'q2']:
#write the output
phrasefile = os.path.join(path, "%s.%s.key" % (q, method))
phrases = fio.LoadList(phrasefile)
netgraphfile = os.path.join(
path,
"%s.%s.%s%s_oslo_files" % (q, method, modelname, net_exe),
'tp')
if not fio.IsExist(netgraphfile): #no communities
print netgraphfile
communites = [[x] for x in range(len(phrases))]
else:
communites = oslom.readgraph_partitions(netgraphfile)
#if len(communites) == 1:#break it
# communites = [[x] for x in range(len(phrases))]
name = 'ct.%s.%s' % (modelname, 'default')
output = os.path.join(
path, "%s.cluster.kmedoids.sqrt.%s.%s" % (q, name, method))
write_communite_to_clusters(communites, phrases, output)
print "%d\t%s\t%d" % (lec, q, len(communites))
def writegraph_leave_one_lecture_out(model_dir,
phrasedir,
modelname='svr',
traincourse=None):
from sklearn import svm
from sklearn.metrics import mean_squared_error, precision_recall_fscore_support, accuracy_score
import QPS_simlearning
sim_extractor = Similarity()
allfeatures = sorted(sim_extractor.features.keys())
features = allfeatures
name = '_'.join(features)
lectures = annotation.Lectures
for i, lec in enumerate(lectures):
test = [lec]
print test
model_file = os.path.join(model_dir, '%d_%s.model' % (lec, name))
#model_file = os.path.join(model_dir, '%s_%s.model'%('IE256_2016', name))
with open(model_file, 'rb') as handle:
clf = pickle.load(handle)
path = os.path.join(phrasedir, str(lec))
for q in ['q1', 'q2']:
test_X, test_Y = QPS_simlearning.combine_files_test(phrasedir,
test,
features,
prompts=[q])
predict_Y = clf.predict(test_X)
#write the output
phrasefile = os.path.join(path, "%s.%s.key" % (q, method))
phrases = fio.LoadList(phrasefile)
assert (len(predict_Y) == len(phrases) * len(phrases))
k = 0
body = []
for i, p1 in enumerate(phrases):
for j, p2 in enumerate(phrases):
if j <= i:
k += 1
continue #undirect graph
if modelname == 'svm':
if predict_Y[k] == 1.0:
#row = [i,j, '%.1f'%predict_Y[k]]
row = [i, j]
body.append(row)
else:
row = [i, j, '%.2f' % predict_Y[k]]
body.append(row)
k += 1
output = os.path.join(
path, "%s.%s.%s%s" % (q, method, modelname, net_exe))
fio.WriteMatrix(output, body)
