Downloads: 1
Diff of
/src/experiments/rf.py
[000000]
..
[735bb5]
Switch to unified view
| a | b/src/experiments/rf.py | ||
|---|---|---|---|
| 1 | #!/usr/bin/env python |
||
| 2 | # -*- coding: utf-8 -*- |
||
| 3 | |||
| 4 | """ |
||
| 5 | Experiments on the Random Forest model and the different datasets (i.e. n2c2, DDI) |
||
| 6 | """ |
||
| 7 | |||
| 8 | # Package Dependencies |
||
| 9 | # -------------------- |
||
| 10 | from .common import final_repetition |
||
| 11 | |||
| 12 | # Local Dependencies |
||
| 13 | # ------------------ |
||
| 14 | from models import RelationCollection |
||
| 15 | from training.base import ALExperimentConfig |
||
| 16 | from training.rf import RandomForestTrainer |
||
| 17 | from training.config import PLExperimentConfig, ALExperimentConfig |
||
| 18 | from utils import set_seed |
||
| 19 | |||
| 20 | # Constants |
||
| 21 | # ---------- |
||
| 22 | from constants import N2C2_REL_TYPES, EXP_RANDOM_SEEDS, RFQueryStrategy |
||
| 23 | |||
| 24 | |||
| 25 | # Experiments |
||
| 26 | # ----------- |
||
| 27 | def rf_passive_learning_n2c2(init_repetiton: int = 0, n_repetitions: int = 5, logging: bool = True): |
||
| 28 | """ |
||
| 29 | Model: Random Forest |
||
| 30 | Dataset: n2c2 |
||
| 31 | Learning: passive |
||
| 32 | """ |
||
| 33 | |||
| 34 | collections = RelationCollection.load_collections("n2c2", splits=["train", "test"]) |
||
| 35 | config = PLExperimentConfig() |
||
| 36 | |||
| 37 | for repetition in range(init_repetiton, final_repetition(init_repetiton, n_repetitions)): |
||
| 38 | # set random seed |
||
| 39 | random_seed = EXP_RANDOM_SEEDS[repetition] |
||
| 40 | set_seed(random_seed) |
||
| 41 | config.seed = random_seed |
||
| 42 | |||
| 43 | for rel_type in N2C2_REL_TYPES: |
||
| 44 | train_collection = collections["train"].type_subcollection(rel_type) |
||
| 45 | test_collection = collections["test"].type_subcollection(rel_type) |
||
| 46 | |||
| 47 | trainer = RandomForestTrainer( |
||
| 48 | dataset="n2c2", |
||
| 49 | train_dataset=train_collection, |
||
| 50 | test_dataset=test_collection, |
||
| 51 | relation_type=rel_type, |
||
| 52 | ) |
||
| 53 | |||
| 54 | trainer.train_passive_learning(config=config, logging=logging) |
||
| 55 | |||
| 56 | |||
| 57 | def rf_passive_learning_ddi(init_repetiton: int = 0, n_repetitions: int = 5, logging: bool = True): |
||
| 58 | """ |
||
| 59 | Model: Random Forest |
||
| 60 | Dataset: DDI |
||
| 61 | Learning: passive |
||
| 62 | """ |
||
| 63 | collections = RelationCollection.load_collections("ddi", splits=["train", "test"]) |
||
| 64 | train_collection = collections["train"] |
||
| 65 | test_collection = collections["test"] |
||
| 66 | config = PLExperimentConfig() |
||
| 67 | |||
| 68 | trainer = RandomForestTrainer( |
||
| 69 | dataset="ddi", |
||
| 70 | train_dataset=train_collection, |
||
| 71 | test_dataset=test_collection, |
||
| 72 | ) |
||
| 73 | |||
| 74 | for repetition in range(init_repetiton, final_repetition(init_repetiton, n_repetitions)): |
||
| 75 | # set random seed |
||
| 76 | random_seed = EXP_RANDOM_SEEDS[repetition] |
||
| 77 | set_seed(random_seed) |
||
| 78 | config.seed = random_seed |
||
| 79 | trainer.train_passive_learning(config=config, logging=logging) |
||
| 80 | |||
| 81 | |||
| 82 | def rf_active_learning_n2c2(init_repetiton: int = 0, n_repetitions: int = 5, logging: bool = True): |
||
| 83 | """ |
||
| 84 | Model: Random Forest |
||
| 85 | Dataset: n2c2 |
||
| 86 | Learning: active |
||
| 87 | """ |
||
| 88 | collections = RelationCollection.load_collections("n2c2", splits=["train", "test"]) |
||
| 89 | config = ALExperimentConfig() |
||
| 90 | |||
| 91 | for repetition in range(init_repetiton, final_repetition(init_repetiton, n_repetitions)): |
||
| 92 | # set random seed |
||
| 93 | random_seed = EXP_RANDOM_SEEDS[repetition] |
||
| 94 | set_seed(random_seed) |
||
| 95 | config.seed = random_seed |
||
| 96 | |||
| 97 | for rel_type in N2C2_REL_TYPES: |
||
| 98 | train_collection = collections["train"].type_subcollection(rel_type) |
||
| 99 | test_collection = collections["test"].type_subcollection(rel_type) |
||
| 100 | |||
| 101 | trainer = RandomForestTrainer( |
||
| 102 | dataset="n2c2", |
||
| 103 | train_dataset=train_collection, |
||
| 104 | test_dataset=test_collection, |
||
| 105 | relation_type=rel_type, |
||
| 106 | ) |
||
| 107 | for query_strategy in RFQueryStrategy: |
||
| 108 | trainer.train_active_learning(query_strategy, config, logging=logging) |
||
| 109 | |||
| 110 | |||
| 111 | def rf_active_learning_ddi(init_repetiton: int = 0, n_repetitions: int = 5, logging: bool = True): |
||
| 112 | """ |
||
| 113 | Model: Random Forest |
||
| 114 | Dataset: DDI |
||
| 115 | Learning: active |
||
| 116 | """ |
||
| 117 | collections = RelationCollection.load_collections("ddi", splits=["train", "test"]) |
||
| 118 | train_collection = collections["train"] |
||
| 119 | test_collection = collections["test"] |
||
| 120 | |||
| 121 | config = ALExperimentConfig() |
||
| 122 | |||
| 123 | for repetition in range(init_repetiton, final_repetition(init_repetiton, n_repetitions)): |
||
| 124 | # set random seed |
||
| 125 | random_seed = EXP_RANDOM_SEEDS[repetition] |
||
| 126 | set_seed(random_seed) |
||
| 127 | config.seed = random_seed |
||
| 128 | |||
| 129 | trainer = RandomForestTrainer( |
||
| 130 | dataset="ddi", |
||
| 131 | train_dataset=train_collection, |
||
| 132 | test_dataset=test_collection, |
||
| 133 | ) |
||
| 134 | for query_strategy in RFQueryStrategy: |
||
| 135 | trainer.train_active_learning(query_strategy, config, logging=logging) |