def asDataFrames(self, *index_by):
'''
Reads the spanned rows as DataFrames if pandas is available, or as
a dict of numpy arrays if only numpy is available or as a dict with
primitives and objects otherwise.
@param index_by If pandas is available, the dataframes will be
indexed by the given columns.
'''
for c in index_by:
if c in self.columns:
raise ValueError(
'column %s cannot be used as index in the data'
'frames as it is a column by which the rows are spanned.')
columns = as_java_array(self.ctx._gateway, "String",
(str(c) for c in self.columns))
jrdd = self._helper.spanBy(self._crdd, columns)
rdd = RDD(jrdd, self.ctx)
global pd
if index_by and pd:
return rdd.mapValues(
lambda _: _.set_index(*[str(c) for c in index_by]))
else:
return rdd
def asDataFrames(self, *index_by):
'''
Reads the spanned rows as DataFrames if pandas is available, or as
a dict of numpy arrays if only numpy is available or as a dict with
primitives and objects otherwise.
@param index_by If pandas is available, the dataframes will be
indexed by the given columns.
'''
for c in index_by:
if c in self.columns:
raise ValueError('column %s cannot be used as index in the data'
'frames as it is a column by which the rows are spanned.')
columns = as_java_array(self.ctx._gateway, "String", (str(c) for c in self.columns))
jrdd = self._helper.spanBy(self._cjrdd, columns)
rdd = RDD(jrdd, self.ctx)
global pd
if index_by and pd:
return rdd.mapValues(lambda _: _.set_index(*[str(c) for c in index_by]))
else:
return rdd
def kmeans(data: RDD,
num_clusters: int,
min_cluster_id=0,
max_iterations=20) -> RDD:
"""
:param data: pair RDD, key is point_id, value is feature vector
:param num_clusters: number of clusters
:param min_cluster_id: The id of each cluster is min_cluster_id, min_cluster_id + 1, min_cluster_id + 2 ...
:param max_iterations: maximum number of iterations
:return: clustering result, pair RDD, key is point_id, value is (cluster id, feature vector)
"""
# ** some helper functions:
def minimum_distance_to_centroids(point: list,
centroids: dict) -> float:
min_distance = min([
sum([(i - j)**2 for i, j in zip(point, centroid)])
for centroid in centroids
])
return min_distance
def assign_point_to_cluster(point: list, centroids_: dict) -> int:
"Given a data point, find the nearest centroid to it"
distances = [(sum([(i - j)**2
for i, j in zip(point, centroid)]), cid)
for cid, centroid in centroids_.items()]
min_distance, cluster_index = min(distances)
return cluster_index
# ** initialize the first k centroids
total_samples = data.count()
print("initializing centroids")
print("total samples:", total_samples)
sample_fraction = 0.3
if total_samples * sample_fraction < num_clusters:
if total_samples < num_clusters:
centroids = data.values().collect()
else:
centroids = data.values().take(num_clusters)
else:
sample_data = data.sample(withReplacement=False,
fraction=sample_fraction,
seed=10)
centroids = [sample_data.first()[1]
] # add first centroid to the collection of centroids
for _ in range(num_clusters - 1): # find the next k-1 centroids
distances = sample_data.\
mapValues(lambda values: minimum_distance_to_centroids(values, centroids)).persist()
furthest_point = sample_data.lookup(
distances.max(lambda pair: pair[1])[0])[0]
centroids.append(
furthest_point) # furthest from already selected points
centroids = [(min_cluster_id + i, centroid)
for i, centroid in enumerate(centroids)] # assign index
centroids = dict(centroids)
# ** perform clustering
num_iterations = 0
previous_cluster_sizes = dict()
while num_iterations < max_iterations:
print("current iteration:", num_iterations)
print("assign points to clusters")
cluster_result: RDD = data.mapValues(lambda values: (
assign_point_to_cluster(values, centroids), values)).persist()
cluster_sizes = cluster_result.map(
lambda pair: (pair[1][0], 1)).reduceByKey(lambda x, y: x + y)
cluster_sizes = dict(cluster_sizes.collect())
if num_iterations > 0: # after first iteration
num_reassigments = [
(cluster_sizes[cid] - previous_cluster_sizes[cid])
for cid in cluster_sizes.keys()
]
print("num_reassigments:", num_reassigments)
if max(num_reassigments) - min(
num_reassigments) < 3: # Now the process converges
break
previous_cluster_sizes = cluster_sizes.copy()
print("update centroids_:")
new_centroids = cluster_result. \
values(). \
flatMapValues(lambda features: [(i, feature) for i, feature in enumerate(features)]). \
map(lambda pair: ((pair[0], pair[1][0]), (pair[1][1], 1))). \
reduceByKey(lambda x, y: (x[0] + y[0], x[1] + y[1])). \
mapValues(lambda values: values[0] / values[1]). \
map(lambda pair: (pair[0][0], (pair[0][1], pair[1]))). \
groupByKey(). \
mapValues(sorted). \
mapValues(lambda values: [value for feature_index, value in values]). \
persist()
new_centroids = dict(new_centroids.collect())
centroids = new_centroids.copy()
num_iterations += 1
print("Converged. Total iterations:", num_iterations)
print(
"cluster size",
cluster_result.values().mapValues(
lambda values: 1).groupByKey().mapValues(len).collect())
return cluster_result
