kaknikhil commented on a change in pull request #433: Kmeans: Add automatic optimal cluster estimation URL: https://github.com/apache/madlib/pull/433#discussion_r325373715
########## File path: src/ports/postgres/modules/kmeans/kmeans_auto.py_in ########## @@ -0,0 +1,226 @@ +# coding=utf-8 +# +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +""" +@file kmeans_auto.py_in + +@brief + +""" + +import numpy as np +import plpy +from utilities.control import MinWarning +from utilities.utilities import _assert +from utilities.utilities import unique_string +from utilities.validate_args import output_tbl_valid +from utilities.validate_args import get_algorithm_name + +ELBOW = 'elbow' +SILHOUETTE = 'silhouette' +BOTH = 'both' + +RANDOM = 'random' +PP = 'pp' + +def _validate(output_table, k): + + output_tbl_valid(output_table, "kmeans_auto") + output_tbl_valid('{0}_summary'.format(output_table), "kmeans_auto") + + _assert(k, "kmeans_auto: k cannot be NULL.") + _assert(len(k)>1, "kmeans_auto: Length of k array should be more than 1.") + _assert(min(k)>1, "kmeans_auto: the minimum k value has to be > 1.") + + k_alt = list(dict.fromkeys(k)) + _assert(len(k) == len(k_alt), "kmeans_auto: Duplicate values are not allowed in k.") + + +def set_defaults(schema_madlib, fn_dist, agg_centroid, max_num_iterations, min_frac_reassigned, k_selection_algorithm, seeding, seeding_sample_ratio): + + fn_dist = (fn_dist if fn_dist else '{0}.squared_dist_norm2'.format(schema_madlib)) + agg_centroid = agg_centroid if agg_centroid \ + else '{0}.avg'.format(schema_madlib) + max_num_iterations = max_num_iterations if max_num_iterations \ + else 20 + min_frac_reassigned = min_frac_reassigned if min_frac_reassigned \ + else 0.001 + + k_selection_algorithm = get_algorithm_name(k_selection_algorithm, SILHOUETTE, + [ELBOW, SILHOUETTE, BOTH], 'kmeans_auto') + + if seeding is PP: + seeding_sample_ratio = (seeding_sample_ratio + if seeding_sample_ratio is not None else 1.0) + return (fn_dist, agg_centroid, max_num_iterations, min_frac_reassigned, + k_selection_algorithm, seeding_sample_ratio) + +def kmeans_auto(schema_madlib, rel_source, output_table, expr_point, k, + fn_dist=None, agg_centroid=None, max_num_iterations=None, + min_frac_reassigned=None, k_selection_algorithm=None, seeding=None, + seeding_sample_ratio=None, **kwargs): + + with MinWarning("error"): + _validate(output_table, k) + + (fn_dist, agg_centroid, max_num_iterations, min_frac_reassigned, + k_selection_algorithm, seeding_sample_ratio) = set_defaults( + schema_madlib, fn_dist, agg_centroid, max_num_iterations, + min_frac_reassigned, k_selection_algorithm, seeding, + seeding_sample_ratio) + + silhouette_col = "" + elbow_col = "" + + # If the selection is elbow or both, calculate elbow + use_silhouette = k_selection_algorithm in [SILHOUETTE, BOTH] + # If the selection is silhouette or both, calculate silhouette + use_elbow = k_selection_algorithm in [ELBOW, BOTH] + + if use_silhouette: + silhouette_col = ", {0} DOUBLE PRECISION".format(SILHOUETTE) + if use_elbow: + elbow_col = ", {0} DOUBLE PRECISION".format(ELBOW) + + plpy.execute(""" + CREATE TABLE {output_table} ( + k INTEGER, + centroids DOUBLE PRECISION[][], + cluster_variance DOUBLE PRECISION[], + objective_fn DOUBLE PRECISION, + frac_reassigned DOUBLE PRECISION, + num_iterations INTEGER + {silhouette_col} + {elbow_col}) + """.format(**locals())) + + silhouette_vals = [] + + for current_k in k: + if seeding is 'random': + plpy.execute(""" + INSERT INTO {output_table} + (k, centroids, cluster_variance, objective_fn, frac_reassigned, + num_iterations) + SELECT {current_k} as k, * + FROM {schema_madlib}.kmeans_random('{rel_source}', + '{expr_point}', + {current_k}, + '{fn_dist}', + '{agg_centroid}', + {max_num_iterations}, + {min_frac_reassigned}); + """.format(**locals())) + else: + plpy.execute(""" + INSERT INTO {output_table} + (k, centroids, cluster_variance, objective_fn, frac_reassigned, + num_iterations) + SELECT {current_k} as k, * + FROM {schema_madlib}.kmeanspp('{rel_source}', + '{expr_point}', + {current_k}, + '{fn_dist}', + '{agg_centroid}', + {max_num_iterations}, + {min_frac_reassigned}, + {seeding_sample_ratio}); + """.format(**locals())) + + if use_silhouette: + silhouette_query= """ + SELECT * FROM {schema_madlib}.simple_silhouette( + '{rel_source}', + '{expr_point}', + (SELECT centroids + FROM {output_table} + WHERE k = {current_k}), + '{fn_dist}') + """.format(**locals()) + silhouette_vals.append( + plpy.execute(silhouette_query)[0]['simple_silhouette']) + + update_query = """ + UPDATE {output_table} SET {{column}} = __value__ FROM + (SELECT unnest(ARRAY[{k_arr}]) AS __k__, + unnest(ARRAY[{{calc_arr}}]) AS __value__ + )sub_q + WHERE __k__ = k + """.format(output_table = output_table, + k_arr = str(k)[1:-1]) + if use_silhouette: + optimal_sil = k[np.argmax(np.array(silhouette_vals))] + plpy.execute(update_query.format(column = SILHOUETTE, + calc_arr = str(silhouette_vals)[1:-1])) + + if use_elbow: + + optimal_elbow, second_order = _calculate_elbow(output_table) + plpy.execute(update_query.format(column = ELBOW, + calc_arr = str(second_order.tolist())[1:-1])) + + optimal_k = optimal_sil if use_silhouette else optimal_elbow + + plpy.execute(""" + CREATE TABLE {output_table}_summary AS + SELECT {output_table}.*, + '{algorithm}'::VARCHAR AS selection_algorithm + FROM {output_table} + WHERE k = {optimal_k} + """.format(algorithm = SILHOUETTE if use_silhouette else ELBOW, + **locals())) + + return + +def _calculate_elbow(output_table): + + # We have to get the values in ordered fashion because the elbow is only defined for ordered values. + inertia_result = plpy.execute(""" + SELECT k, objective_fn FROM {output_table} ORDER BY k ASC + """.format(**locals())) + k = [ i['k'] for i in inertia_result ] + inertia_list = [ i['objective_fn'] for i in inertia_result ] + inertia_list = np.array(inertia_list) + + first_order=np.gradient(inertia_list, k) + second_order=np.gradient(first_order, k) + index_with_elbow=k[np.argmax(second_order)] + + return index_with_elbow, second_order Review comment: we should return second_order.tolist() since that is how it gets used ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
