def cv(data, probs):
cluster_cnt = data.cluster_cnt * 3
return Pipe() \
.x(data.X) \
.y(data.Y) \
.pipe(badness_denclue(bandwidth=data.bandwidth, prepare=True)) \
.split(len(probs), seeder(probs)) \
.pipe(badness_denclue()) \
.split(10, cross('y_seed')) \
.connect(kmeans_ssl(cluster_cnt, data.K_for_KNN)) \
.merge('evaluation', total('evaluation')) \
.merge('result', group('evaluation', 'badness_denclue', 'name')) \
.connect(stop())
def normal(data, probs):
cluster_cnt = data.cluster_cnt * 3
return Pipe() \
.x(data.X) \
.y(data.Y) \
.x_test(data.X_test) \
.y_test(data.Y_test) \
.pipe(badness_agglomeratvie_l_method(prepare=True)) \
.split(len(probs), seeder(probs))\
.pipe(badness_agglomeratvie_l_method()) \
.connect(kmeans_ssl(cluster_cnt, data.K_for_KNN)) \
.merge('result', group('evaluation', 'badness', 'name'))\
.connect(stop())
def cv(data, prob):
cluster_cnt = data.cluster_cnt * 3
return Pipe() \
.x(data.X) \
.y(data.Y) \
.pipe(badness_agglomeratvie_l_method(prepare=True)) \
.split(random_cnt) \
.y_seed(seeding_random(prob)) \
.pipe(badness_agglomeratvie_l_method()) \
.split(10, cross('y_seed')) \
.connect(kmeans_ssl(cluster_cnt, data.K_for_KNN)) \
.merge('evaluation', total('evaluation')) \
.merge('result', group('evaluation', 'badness')) \
.connect(stop())
def kmeans_ssl(clusters, neighbors):
def fn(pipe):
p = pipe \
.pipe(kmeans(clusters)) \
.y(label_consensus()) \
.pipe(knn(neighbors)) \
.pipe(predict()) \
.pipe(evaluate())
return p
return fn
p = Pipe() \
.x(data.X) \
.y(data.Y) \
.x_test(data.X_test) \
.y_test(data.Y_test) \
.pipe(badness_agglomeratvie_l_method(prepare=True))\
.split(3)\
.y_seed(seeding_random(0.1)) \
.pipe(badness_agglomeratvie_l_method()) \
.connect(kmeans_ssl(clusters_count, data.K_for_KNN))\
.merge('result', group('evaluation', 'badness'))\
.connect(stop())
result = p['result']
print('result:', result)
with open('results/badness_agglomerative_l_method-' + data.name + '.json', 'w') as file:
json.dump(result, file)