| #------------------------------------------------------------- |
| # |
| # 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. |
| # |
| #------------------------------------------------------------- |
| |
| from slicing.base.node import Node |
| from slicing.base.top_k import Topk |
| |
| |
| # optimization function calculation: |
| # fi - error of subset, si - its size |
| # f - error of complete set, x_size - its size |
| # w - "significance" weight of error constraint |
| def opt_fun(fi, si, f, x_size, w): |
| formula = w * (fi/f) + (1 - w) * (si/x_size) |
| return float(formula) |
| |
| |
| # slice_name_nonsense function defines if combination of nodes on current level is fine or impossible: |
| # there is dependency between common nodes' attributes number and current level is such that: |
| # commons == cur_lvl - 1 |
| # valid combination example: node ABC + node BCD (on 4th level) // three attributes nodes have two common attributes |
| # invalid combination example: node ABC + CDE (on 4th level) // result node - ABCDE (absurd for 4th level) |
| def slice_name_nonsense(node_i, node_j, cur_lvl): |
| attr1 = list(map(lambda x: x[0].split("_")[0], node_i.attributes)) |
| attr2 = list(map(lambda x: x[0].split("_")[0], node_j.attributes)) |
| commons = len(list(set(attr1) & set(attr2))) |
| return commons == cur_lvl - 1 |
| |
| |
| def union(lst1, lst2): |
| final_list = sorted(set(list(lst1) + list(lst2))) |
| return final_list |
| |
| |
| def make_first_level(all_features, complete_x, loss, x_size, y_test, errors, loss_type, top_k, alpha, w): |
| first_level = [] |
| counter = 0 |
| all_nodes = {} |
| # First level slices are enumerated in a "classic way" (getting data and not analyzing bounds |
| for feature in all_features: |
| new_node = Node(complete_x, loss, x_size, y_test, errors) |
| new_node.parents = [(feature, counter)] |
| new_node.attributes.append((feature, counter)) |
| new_node.name = new_node.make_name() |
| new_id = len(all_nodes) |
| new_node.key = new_node.make_key(new_id) |
| all_nodes[new_node.key] = new_node |
| new_node.process_slice(loss_type) |
| new_node.score = opt_fun(new_node.loss, new_node.size, loss, x_size, w) |
| new_node.c_upper = new_node.score |
| first_level.append(new_node) |
| new_node.print_debug(top_k, 0) |
| # constraints for 1st level nodes to be problematic candidates |
| if new_node.check_constraint(top_k, x_size, alpha): |
| # this method updates top k slices if needed |
| top_k.add_new_top_slice(new_node) |
| counter = counter + 1 |
| return first_level, all_nodes |
| |
| |
| def join_enum(node_i, prev_lvl, complete_x, loss, x_size, y_test, errors, debug, alpha, w, loss_type, b_update, cur_lvl, |
| all_nodes, top_k, cur_lvl_nodes): |
| for node_j in range(len(prev_lvl)): |
| flag = slice_name_nonsense(prev_lvl[node_i], prev_lvl[node_j], cur_lvl) |
| if flag and prev_lvl[node_j].key[0] > prev_lvl[node_i].key[0]: |
| new_node = Node(complete_x, loss, x_size, y_test, errors) |
| parents_set = set(new_node.parents) |
| parents_set.add(prev_lvl[node_i]) |
| parents_set.add(prev_lvl[node_j]) |
| new_node.parents = list(parents_set) |
| parent1_attr = prev_lvl[node_i].attributes |
| parent2_attr = prev_lvl[node_j].attributes |
| new_node_attr = union(parent1_attr, parent2_attr) |
| new_node.attributes = new_node_attr |
| new_node.name = new_node.make_name() |
| new_id = len(all_nodes) |
| new_node.key = new_node.make_key(new_id) |
| if new_node.key[1] in all_nodes: |
| existing_item = all_nodes[new_node.key[1]] |
| parents_set = set(existing_item.parents) |
| existing_item.parents = parents_set |
| if b_update: |
| s_upper = new_node.calc_s_upper(cur_lvl) |
| s_lower = new_node.calc_s_lower(cur_lvl) |
| e_upper = new_node.calc_e_upper() |
| e_max_upper = new_node.calc_e_max_upper(cur_lvl) |
| new_node.update_bounds(s_upper, s_lower, e_upper, e_max_upper, w) |
| else: |
| new_node.calc_bounds(cur_lvl, w) |
| all_nodes[new_node.key[1]] = new_node |
| # check if concrete data should be extracted or not (only for those that have score upper |
| # big enough and if size of subset is big enough |
| to_slice = new_node.check_bounds(top_k, x_size, alpha) |
| if to_slice: |
| new_node.process_slice(loss_type) |
| new_node.score = opt_fun(new_node.loss, new_node.size, loss, x_size, w) |
| # we decide to add node to current level nodes (in order to make new combinations |
| # on the next one or not basing on its score value |
| if new_node.check_constraint(top_k, x_size, alpha) and new_node.key not in top_k.keys: |
| top_k.add_new_top_slice(new_node) |
| cur_lvl_nodes.append(new_node) |
| if debug: |
| new_node.print_debug(top_k, cur_lvl) |
| return cur_lvl_nodes, all_nodes |
| |
| |
| # alpha is size significance coefficient (required for optimization function) |
| # verbose option is for returning debug info while creating slices and printing it (in console) |
| # k is number of top-slices we want to receive as a result (maximum output, if less all of subsets will be printed) |
| # w is a weight of error function significance (1 - w) is a size significance propagated into optimization function |
| # loss_type = 0 (in case of regression model) |
| # loss_type = 1 (in case of classification model) |
| def process(all_features, complete_x, loss, x_size, y_test, errors, debug, alpha, k, w, loss_type, b_update): |
| levels = [] |
| top_k = Topk(k) |
| first_level = make_first_level(all_features, complete_x, loss, x_size, y_test, errors, loss_type, top_k, alpha, w) |
| all_nodes = first_level[1] |
| levels.append(first_level[0]) |
| # cur_lvl - index of current level, correlates with number of slice forming features |
| cur_lvl = 1 # currently filled level after first init iteration |
| # currently for debug |
| print("Level 1 had " + str(len(all_features)) + " candidates") |
| print() |
| print("Current topk are: ") |
| top_k.print_topk() |
| # combining each candidate of previous level with every till it becomes useless (one node can't make a pair) |
| while len(levels[cur_lvl - 1]) > 1: |
| cur_lvl_nodes = [] |
| prev_lvl = levels[cur_lvl - 1] |
| for node_i in range(len(prev_lvl)): |
| partial = join_enum(node_i, prev_lvl, complete_x, loss, x_size, y_test, errors, debug, alpha, w, loss_type, |
| b_update, cur_lvl, all_nodes, top_k, cur_lvl_nodes) |
| cur_lvl_nodes = partial[0] |
| all_nodes = partial[1] |
| cur_lvl = cur_lvl + 1 |
| levels.append(cur_lvl_nodes) |
| top_k.print_topk() |
| print("Level " + str(cur_lvl) + " had " + str(len(prev_lvl) * (len(prev_lvl) - 1)) + |
| " candidates but after pruning only " + str(len(cur_lvl_nodes)) + " go to the next level") |
| print("Program stopped at level " + str(cur_lvl + 1)) |
| print() |
| print("Selected slices are: ") |
| top_k.print_topk() |
