def output(partId, ch_aux):
"""Uses the student code to compute the output for test cases."""
version = 1
output = [partId, version]
irsys = IRSystem()
irsys.read_data('../data/RiderHaggard')
irsys.index()
irsys.compute_tfidf()
out = sys.stdout
if partId in [2,4,6,8,10]: # test parts
sys.stdout = open(os.devnull, 'w')
if partId == 1 or partId == 2:
# Inverted Index. 1 ==> dev; 2 ==> test
queries = ch_aux.split(", ")
for query in queries:
posting = irsys.get_posting_unstemmed(query)
output.append(list(posting))
elif partId == 3 or partId == 4:
# Boolean Retrieval. 3 ==> dev; 4 ==> test
queries = ch_aux.split(", ")
for query in queries:
result = irsys.query_retrieve(query)
result = list(result)
output.append(result)
elif partId == 5 or partId == 6:
# Phrase Query Retrieval. 5 ==> dev; 6 ==> test
queries = ch_aux.split(",")
print queries
for query in queries:
result = irsys.phrase_query_retrieve(query)
result = list(result)
output.append(result)
elif partId == 7 or partId == 8:
# TF-IDF. 7 ==> dev; 8 ==> test
queries = ch_aux.split("; ")
for query in queries:
word, docID = query.split(", ")
result = irsys.get_tfidf_unstemmed(word, int(docID));
#print 'word: "%s" docID: "%s" result: %f
output.append(result)
elif partId == 9 or partId == 10:
# Cosine Similarity. 9 ==> dev; 10 ==> test
queries = ch_aux.split(", ")
for query in queries:
results = irsys.query_rank(query)
first_result = [results[0][0], results[0][1]]
output.append(first_result)
else:
print "Unknown partId: %d" % partId
return None
if partId in [2,4,6,8,10]: # test parts
sys.stdout = out
# put in the part ID as well (hacky)
output = str(output)
#print 'output: %s' % output
return output
def output(self, partId, ch_aux):
"""Uses the student code to compute the output for test cases."""
version = 1
output = [partId, version]
irsys = IRSystem()
irsys.read_data('../data/RiderHaggard')
irsys.index()
irsys.compute_tfidf()
out = sys.stdout
if partId in [2,4,6,8,10]: # test parts
sys.stdout = open(os.devnull, 'w')
if partId == 1 or partId == 2:
# Inverted Index. 1 ==> dev; 2 ==> test
queries = ch_aux.split(", ")
for query in queries:
posting = irsys.get_posting_unstemmed(query)
output.append(list(posting))
elif partId == 3 or partId == 4:
# Boolean Retrieval. 3 ==> dev; 4 ==> test
queries = ch_aux.split(", ")
for query in queries:
result = irsys.query_retrieve(query)
result = list(result)
output.append(result)
elif partId == 5 or partId == 6:
# Phrase Query Retrieval. 5 ==> dev; 6 ==> test
queries = ch_aux.split(",")
print queries
for query in queries:
result = irsys.phrase_query_retrieve(query)
result = list(result)
output.append(result)
elif partId == 7 or partId == 8:
# TF-IDF. 7 ==> dev; 8 ==> test
queries = ch_aux.split("; ")
for query in queries:
word, docID = query.split(", ")
result = irsys.get_tfidf_unstemmed(word, int(docID));
#print 'word: "%s" docID: "%s" result: %f
output.append(result)
elif partId == 9 or partId == 10:
# Cosine Similarity. 9 ==> dev; 10 ==> test
queries = ch_aux.split(", ")
for query in queries:
results = irsys.query_rank(query)
first_result = [results[0][0], results[0][1]]
output.append(first_result)
else:
print "Unknown partId: %d" % partId
return None
if partId in [2,4,6,8,10]: # test parts
sys.stdout = out
# put in the part ID as well (hacky)
output = str(output)
#print 'output: %s' % output
return output
if __name__ == '__main__':
for labelfile in ("labels.hcnorms_misgoodfortune.all.txt",
"labels.hcnorms_char.all.txt",
"dream_acts.txt", 'dream_sets.txt', "all_labels.txt"):
# First we'll do a regular IR experiment with BM25
documents = {doc_id: text for doc_id, text in read_dreams("data/dreambank.en.stanford.out")}
labels = list(read_labels("data/" + labelfile))
y, X = zip(*match_labels_documents(documents, labels))
y, X = np.array(y), np.array(X)
kf = KFold(len(y), n_folds=10, shuffle=True, random_state=1)
rank_scores = np.zeros(10)
for i, (train, test) in enumerate(kf):
X_train, X_test, y_train, y_test = X[train], X[test], y[train], y[test]
labels = Counter(flatten(list(y_train)))
labels = [label for label, count in labels.items() if count >= 1]
model = IRSystem(k1=1.2, b=0.75, cutoff=0)
model.fit_raw(X_train, y_train, ngram_range=(1, 1), stop_words='english', min_df=2)
ranking = model.rank_labels(X_test, raw=True)
ranking = ranking.tolist()
ranking = map(lambda r: list(unique_everseen(r)), map(flatten, ranking))
ranking, y_test = zip(*[(r, y_) for r, y_ in zip(ranking, y_test) if any(l in labels for l in y_)])
rank_scores[i] = mean_average_precision(ranking, y_test)
print 'IR: (%s)' % (labelfile), rank_scores.mean(), rank_scores.std()
# Next, we'll do an IR experiment with Big Documents
documents = {doc_id: text for doc_id, text in read_dreams("data/dreambank.en.stanford.out")}
labels = list(read_labels("data/" + labelfile))
y, X = zip(*match_labels_documents(documents, labels))
y, X = np.array(y), np.array(X)
kf = KFold(len(y), n_folds=10, shuffle=True, random_state=1)
rank_scores = np.zeros(10)
