Source code for asreview.query_strategies.cluster

# Copyright 2019 The ASReview Authors. All Rights Reserved.
#
# Licensed 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.

import numpy as np
from sklearn.cluster import KMeans

from asreview.query_strategies.base import ProbaQueryStrategy
from asreview.query_strategies.max import MaxQuery
from asreview.utils import get_random_state


[docs]class ClusterQuery(ProbaQueryStrategy): "Query strategy using clustering algorithms." name = "cluster" def __init__(self, cluster_size=350, update_interval=200, random_state=None): """Initialize the clustering strategy. Arguments --------- cluster_size: int Size of the clusters to be made. If the size of the clusters is smaller than the size of the pool, fall back to max sampling. update_interval: int Update the clustering every x instances. random_state: int, RandomState State/seed of the RNG. """ super(ClusterQuery, self).__init__() self.cluster_size = cluster_size self.update_interval = update_interval self.last_update = None self.fallback_model = MaxQuery() self._random_state = get_random_state(random_state) def _query(self, X, pool_idx, n_instances, proba): n_samples = X.shape[0] if pool_idx is None: pool_idx = np.arange(n_samples) last_update = self.last_update if (last_update is None or self.update_interval is None or last_update-len(pool_idx) >= self.update_interval): n_clusters = round(len(pool_idx)/self.cluster_size) if n_clusters <= 1: return self.fallback_model._query( X, pool_idx=pool_idx, n_instances=n_instances, proba=proba) model = KMeans(n_clusters=n_clusters, n_init=1, random_state=self._random_state) self.clusters = model.fit_predict(X) self.last_update = len(pool_idx) clusters = {} for idx in pool_idx: cluster_id = self.clusters[idx] if cluster_id in clusters: clusters[cluster_id].append((idx, proba[idx, 1])) else: clusters[cluster_id] = [(idx, proba[idx, 1])] for cluster_id in clusters: try: clusters[cluster_id] = sorted( clusters[cluster_id], key=lambda x: x[1]) except ValueError: raise clust_idx = [] cluster_ids = list(clusters) for _ in range(n_instances): cluster_id = self._random_state.choice(cluster_ids, 1)[0] clust_idx.append(clusters[cluster_id].pop()[0]) if len(clusters[cluster_id]) == 0: del clusters[cluster_id] cluster_ids = list(clusters) clust_idx = np.array(clust_idx, dtype=int) return clust_idx, X[clust_idx] def full_hyper_space(self): from hyperopt import hp parameter_space = { "qry_cluster_size": hp.quniform('qry_cluster_size', 50, 1000, 1), "qry_update_interval": hp.quniform( 'qry_update_interval', 100, 300, 1), } return parameter_space, {}