def setUp(self):
self.points = np.array([[2.5, 0.1], [2, 0.2], [3, 0.1], [4, 0.2],
[0.1, 2.5], [0.2, 2], [0.1, 3], [0.2, 4]],
dtype=np.float32)
self.num_points = self.points.shape[0]
self.true_centers = np.array([
normalize(
np.mean(normalize(self.points)[0:4, :], axis=0,
keepdims=True))[0],
normalize(
np.mean(normalize(self.points)[4:, :], axis=0,
keepdims=True))[0]
],
dtype=np.float32)
self.true_assignments = np.array([0] * 4 + [1] * 4)
self.true_score = len(self.points) - np.tensordot(
normalize(self.points), self.true_centers[self.true_assignments])
self.num_centers = 2
self.kmeans = kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=factorization.RANDOM_INIT,
distance_metric=factorization.COSINE_DISTANCE,
use_mini_batch=self.use_mini_batch,
mini_batch_steps_per_iteration=self.mini_batch_steps_per_iteration,
config=self.config(3))
def main(_):
train_set = np.load(FLAGS.train_file)
val_set = np.load(FLAGS.test_file)
num_k = int(FLAGS.num_k)
clusterer = kmeans.KMeansClustering(
num_clusters=num_k,
use_mini_batch=False)
labels = train_set[:,0:1]
if FLAGS.endtoend:
ks = train_set[:1:2]
x = train_set[:,2:]
else:
x = train_set[:,1:]
clusterer.fit(
input_fn=_input_fn(x.astype(np.float32)), steps=10)
labels_val = val_set[:,0:1]
if FLAGS.endtoend:
ks_val = val_set[:,1:2]
x_val = val_set[:,2:]
else:
x_val = val_set[:,1:]
predictions = np.array(list(clusterer.predict_cluster_idx(
input_fn=_input_fn(x_val.astype(np.float32), num_epochs=1))))
process_clusters('K Cluster', labels_val.flatten(), predictions)
if FLAGS.endtoend:
process_clusters('K end-to-end', labels_val.flatten(), ks_val.flatten().astype(int))
def create_experiment_fn(output_dir=None):
"""Experiment function."""
distance_metric = (tf.contrib.factorization.COSINE_DISTANCE
if FLAGS.use_cosine_distance else
tf.contrib.factorization.SQUARED_EUCLIDEAN_DISTANCE)
initial_clusters = (tf.contrib.factorization.KMEANS_PLUS_PLUS_INIT
if FLAGS.use_kmeans_plus_plus else
tf.contrib.factorization.RANDOM_INIT)
# Create estimator
kmeans = kmeans_lib.KMeansClustering(
FLAGS.num_clusters,
model_dir=output_dir,
initial_clusters=initial_clusters,
distance_metric=distance_metric,
use_mini_batch=True,
relative_tolerance=FLAGS.relative_tolerance,
config=tf.contrib.learn.RunConfig(
save_checkpoints_secs=FLAGS.save_checkpoints_secs))
train_monitors = []
if FLAGS.debug:
train_monitors.append(tf_debug.LocalCLIDebugHook())
return tf.contrib.learn.Experiment(
estimator=kmeans,
train_steps=FLAGS.num_train_steps,
eval_steps=1,
eval_input_fn=_input_fn,
train_input_fn=_input_fn,
train_monitors=train_monitors,
export_strategies=[
saved_model_export_utils.make_export_strategy(_predict_input_fn,
exports_to_keep=5)
])
def main(_):
filelist = []
for line in open(FLAGS.file_list):
filelist.append(line)
train_set = np.load(FLAGS.train_file)
val_set = np.load(FLAGS.test_file)
num_k = int(FLAGS.num_k)
clusterer = kmeans.KMeansClustering(
num_clusters=num_k,
use_mini_batch=False)
labels = train_set[:,0:1]
if FLAGS.endtoend:
ks = train_set[:1:2]
x = train_set[:,2:]
else:
x = train_set[:,1:]
clusterer.fit(
input_fn=_input_fn(x.astype(np.float32)), steps=10)
labels_val = val_set[:,0:1]
if FLAGS.endtoend:
ks_val = val_set[:,1:2]
x_val = val_set[:,2:]
else:
x_val = val_set[:,1:]
predictions = np.array(list(clusterer.predict_cluster_idx(
input_fn=_input_fn(x_val.astype(np.float32), num_epochs=1))))
file_dict = dict()
c = 0
for f in filelist:
f = f.strip()
fields = f.split(',')
if int(fields[3]) == 1:
# See if field[1] exists in dict
key = fields[1]
info = fields[0] + ',' + str(predictions[c]) + ',' + fields[4] + ',' + fields[5] + ',' + fields[6] + ',' + fields[7]
if key not in file_dict:
file_dict[key] = []
file_dict[key].append(info)
c += 1
colors = [(255,0,0),(0,255,0),(0,0,255),(255,255,0)]
for k,v in file_dict.iteritems():
fields = k.strip().split('/')
filename = os.path.join(FLAGS.output_folder,fields[-1].replace('leftImg8bit','results'))
img = cv2.imread(k.strip())
for info in v:
fields = info.split(',')
label = int(fields[1])
x = int(fields[2])
y = int(fields[3])
w = int(fields[4])
h = int(fields[5])
cv2.rectangle(img,(x,y),(x+w,y+h),colors[label],2)
cv2.imwrite(filename, img)
def _kmeans(self, relative_tolerance=None):
return kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=factorization.RANDOM_INIT,
use_mini_batch=self.use_mini_batch,
config=self.config(14),
random_seed=10,
relative_tolerance=relative_tolerance)
def _kmeans(self, relative_tolerance=None):
return kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=factorization.KMEANS_PLUS_PLUS_INIT,
distance_metric=factorization.SQUARED_EUCLIDEAN_DISTANCE,
use_mini_batch=self.use_mini_batch,
mini_batch_steps_per_iteration=self.mini_batch_steps_per_iteration,
random_seed=24,
relative_tolerance=relative_tolerance)
def test_fit_raise_if_num_clusters_larger_than_num_points_random_init(
self):
points = np.array([[2.0, 3.0], [1.6, 8.2]], dtype=np.float32)
with self.assertRaisesOpError('less'):
kmeans = kmeans_lib.KMeansClustering(
num_clusters=3, initial_clusters=factorization.RANDOM_INIT)
kmeans.fit(input_fn=lambda: (constant_op.constant(points), None),
steps=10)
def generate_cluster(k, data_set):
"""
This will generate centroids of k cluster from the given date set
:param k number of clusters to generate
:param data_set input data set
"""
k_means_estimator = kmeans.KMeansClustering(num_clusters=k)
k_means_estimator.fit(input_fn=lambda: input_fn_1d(data_set), steps=1000)
return k_means_estimator.clusters()
def test_kmeans_plus_plus_batch_too_small(self):
points = np.array(
[[1, 2], [3, 4], [5, 6], [7, 8], [9, 0]], dtype=np.float32)
kmeans = kmeans_lib.KMeansClustering(
num_clusters=points.shape[0],
initial_clusters=kmeans_lib.KMeansClustering.KMEANS_PLUS_PLUS_INIT,
distance_metric=kmeans_lib.KMeansClustering.SQUARED_EUCLIDEAN_DISTANCE,
use_mini_batch=True,
mini_batch_steps_per_iteration=100,
random_seed=24,
relative_tolerance=None)
with self.assertRaisesOpError(AssertionError):
kmeans.fit(
input_fn=self.input_fn(batch_size=4, points=points, randomize=False),
steps=1)
def test_kmeans_plus_plus_batch_just_right(self):
points = np.array([[1, 2]], dtype=np.float32)
kmeans = kmeans_lib.KMeansClustering(
num_clusters=points.shape[0],
initial_clusters=kmeans_lib.KMeansClustering.KMEANS_PLUS_PLUS_INIT,
distance_metric=kmeans_lib.KMeansClustering.SQUARED_EUCLIDEAN_DISTANCE,
use_mini_batch=True,
mini_batch_steps_per_iteration=100,
random_seed=24,
relative_tolerance=None)
kmeans.fit(
input_fn=self.input_fn(batch_size=1, points=points, randomize=False),
steps=1)
clusters = kmeans.clusters()
self.assertAllEqual(points, clusters)
def test_predict_kmeans_plus_plus(self):
# Most points are concetrated near one center. KMeans++ is likely to find
# the less populated centers.
points = np.array([[2.5, 3.5], [2.5, 3.5], [-2, 3], [-2, 3], [-3, -3],
[-3.1, -3.2], [-2.8, -3.], [-2.9, -3.1],
[-3., -3.1], [-3., -3.1], [-3.2, -3.], [-3., -3.]],
dtype=np.float32)
true_centers = np.array([
normalize(np.mean(normalize(points)[0:2, :], axis=0,
keepdims=True))[0],
normalize(np.mean(normalize(points)[2:4, :], axis=0,
keepdims=True))[0],
normalize(np.mean(normalize(points)[4:, :], axis=0,
keepdims=True))[0]
],
dtype=np.float32)
true_assignments = [0] * 2 + [1] * 2 + [2] * 8
true_score = len(points) - np.tensordot(normalize(points),
true_centers[true_assignments])
kmeans = kmeans_lib.KMeansClustering(
3,
initial_clusters=factorization.KMEANS_PLUS_PLUS_INIT,
distance_metric=factorization.COSINE_DISTANCE,
use_mini_batch=self.use_mini_batch,
mini_batch_steps_per_iteration=self.mini_batch_steps_per_iteration,
config=self.config(3))
kmeans.fit(input_fn=lambda: (constant_op.constant(points), None),
steps=30)
centers = normalize(kmeans.clusters())
self.assertAllClose(sorted(centers.tolist()),
sorted(true_centers.tolist()),
atol=1e-2)
def _input_fn():
return (input_lib.limit_epochs(constant_op.constant(points),
num_epochs=1), None)
assignments = list(kmeans.predict_cluster_idx(input_fn=_input_fn))
self.assertAllClose(centers[assignments],
true_centers[true_assignments],
atol=1e-2)
score = kmeans.score(input_fn=lambda:
(constant_op.constant(points), None),
steps=1)
self.assertAllClose(score, true_score, atol=1e-2)
def setUp(self):
np.random.seed(3)
random_seed_lib.set_random_seed(2)
self.num_centers = 2
self.num_dims = 2
self.num_points = 4000
self.batch_size = self.num_points
self.true_centers = self.make_random_centers(self.num_centers,
self.num_dims)
self.points, self.assignments = self.make_random_points(
self.true_centers, self.num_points)
# Use initial means from kmeans (just like scikit-learn does).
clusterer = kmeans.KMeansClustering(num_clusters=self.num_centers)
clusterer.fit(input_fn=lambda: (constant_op.constant(self.points), None),
steps=30)
self.initial_means = clusterer.clusters()
def setUp(self):
np.random.seed(3)
self.num_centers = 5
self.num_dims = 2
self.num_points = 10000
self.true_centers = make_random_centers(self.num_centers,
self.num_dims)
self.points, _, self.scores = make_random_points(
self.true_centers, self.num_points)
self.true_score = np.add.reduce(self.scores)
self.kmeans = kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=factorization.RANDOM_INIT,
use_mini_batch=self.use_mini_batch,
config=self.config(14),
random_seed=10)
def test_monitor(self):
if self.use_mini_batch:
return
kmeans = kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=factorization.RANDOM_INIT,
use_mini_batch=self.use_mini_batch,
config=run_config.RunConfig(tf_random_seed=14),
random_seed=12)
kmeans.fit(
input_fn=self.input_fn(),
# Force it to train forever until the monitor stops it.
steps=None,
relative_tolerance=1e-4)
score = kmeans.score(
input_fn=self.input_fn(batch_size=self.num_points), steps=1)
self.assertNear(self.true_score, score, self.true_score * 0.005)
def _fit(self, num_iters=10):
scores = []
start = time.time()
for i in range(num_iters):
print('Starting tensorflow KMeans: %d' % i)
tf_kmeans = kmeans_lib.KMeansClustering(
self.num_clusters,
initial_clusters=kmeans_lib.KMeansClustering.KMEANS_PLUS_PLUS_INIT,
kmeans_plus_plus_num_retries=int(math.log(self.num_clusters) + 2),
random_seed=i * 42,
relative_tolerance=1e-6,
config=run_config.RunConfig(tf_random_seed=3))
tf_kmeans.fit(
input_fn=lambda: (constant_op.constant(self.points), None), steps=50)
_ = tf_kmeans.clusters()
scores.append(
tf_kmeans.score(
input_fn=lambda: (constant_op.constant(self.points), None),
steps=1))
self._report(num_iters, start, time.time(), scores)
def test_monitor(self):
if self.use_mini_batch:
# We don't test for use_mini_batch case since the loss value can be noisy.
return
kmeans = kmeans_lib.KMeansClustering(
self.num_centers,
initial_clusters=kmeans_lib.KMeansClustering.KMEANS_PLUS_PLUS_INIT,
distance_metric=kmeans_lib.KMeansClustering.SQUARED_EUCLIDEAN_DISTANCE,
use_mini_batch=self.use_mini_batch,
mini_batch_steps_per_iteration=self.mini_batch_steps_per_iteration,
config=learn.RunConfig(tf_random_seed=14),
random_seed=12,
relative_tolerance=1e-4)
kmeans.fit(
input_fn=self.input_fn(),
# Force it to train until the relative tolerance monitor stops it.
steps=None)
score = kmeans.score(
input_fn=self.input_fn(batch_size=self.num_points), steps=1)
def test_queues(self):
kmeans = kmeans_lib.KMeansClustering(5)
kmeans.fit(input_fn=self.input_fn(), steps=1)
# __init__(
# num_clusters,
# model_dir=None,
# initial_clusters=RANDOM_INIT,
# distance_metric=SQUARED_EUCLIDEAN_DISTANCE,
# random_seed=0,
# use_mini_batch=True,
# mini_batch_steps_per_iteration=1,
# kmeans_plus_plus_num_retries=2,
# relative_tolerance=None,
# config=None
# )
# In[16]:
k_means_estimator = kmeans.KMeansClustering(num_clusters=10)
# Start with a 1000 steps and then try 10000
fit = k_means_estimator.fit(input_fn=lambda: input_fn(training_digits),
steps=1000)
# In[17]:
clusters = k_means_estimator.clusters()
# ### Plot the image representations of the centroids of the clusters
# *Note that these may not be actual images in our training data*
# In[20]:
for i in range(10):
