/
util.py
218 lines (182 loc) · 9.9 KB
/
util.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
import re
from collections import OrderedDict
import numpy as np
from keras.losses import categorical_crossentropy
from keras.metrics import categorical_accuracy
from keras.optimizers import SGD
from skopt import load
from csrank.callbacks import DebugOutput, LRScheduler
from csrank.constants import OBJECT_RANKING, LABEL_RANKING, DYAD_RANKING, DISCRETE_CHOICE, BATCH_SIZE, LEARNING_RATE, \
LOG_UNIFORM
from csrank.dataset_reader import DepthDatasetReader, ImageDatasetReader, SushiObjectRankingDatasetReader, \
SyntheticDatasetGenerator, TagGenomeDatasetReader, SentenceOrderingDatasetReader, LetorObjectRankingDatasetReader
from csrank.fate_ranking import FATEObjectRanker, N_HIDDEN_SET_LAYERS, N_HIDDEN_JOINT_LAYERS, \
N_HIDDEN_SET_UNITS, N_HIDDEN_JOINT_UNITS
from csrank.losses import smooth_rank_loss
from csrank.metrics import zero_one_rank_loss_for_scores_ties, zero_one_rank_loss_for_scores, \
spearman_correlation_for_scores, kendalls_tau_for_scores, zero_one_accuracy_for_scores
from csrank.objectranking.feta_ranker import FETANetwork
from csrank.objectranking.cmp_net import *
from csrank.objectranking.expected_rank_regression import ExpectedRankRegression
from csrank.objectranking.rank_net import RankNet
from csrank.objectranking.rank_svm import RankSVM
from csrank.util import kendalls_mean_np, spearman_mean_np, zero_one_accuracy_np, \
zero_one_rank_loss_for_scores_ties_np, zero_one_rank_loss_for_scores_np
HYPER_VOLUME = "hyper_volume"
DEPTH = 'depth'
SENTENCE_ORDERING = "sentence_ordering"
LETOR = "letor"
BORDA_RANKNET = "bordaranknet"
BORDA_RANKNET_ZERO = "bordaranknet_zero"
RANKSVM = 'ranksvm'
ERR = 'err'
CMPNET = "cmpnet"
RANKNET = 'ranknet'
BORDA_ZERO = 'borda_zero'
BORDA = 'borda'
GENERAL_RANKER = "general_ranker"
object_ranking_datasets = {'medoid': SyntheticDatasetGenerator, HYPER_VOLUME: SyntheticDatasetGenerator,
DEPTH: DepthDatasetReader,
'sushi': SushiObjectRankingDatasetReader, 'image_dataset': ImageDatasetReader,
'tag_genome': TagGenomeDatasetReader,
SENTENCE_ORDERING: SentenceOrderingDatasetReader,
LETOR: LetorObjectRankingDatasetReader} # 'depth_semantic': generate_depth_dataset, 'depth_basic': generate_depth_dataset
dataset_arg_options = {HYPER_VOLUME: None, 'medoid': None, DEPTH: "dataset_type",
'sushi': None, 'image_dataset': None,
'tag_genome': None,
SENTENCE_ORDERING: ["n_dims", "train_obj"], LETOR: ["year", "train_obj"]}
object_rankers = {BORDA: FETANetwork, BORDA_ZERO: FETANetwork, RANKNET: RankNet, CMPNET: CmpNet,
ERR: ExpectedRankRegression, RANKSVM: RankSVM,
GENERAL_RANKER: FATEObjectRanker} # , "bordaranknet_zero", "bordaranknet"
ranking_metrics = OrderedDict(
{'KendallsTau': kendalls_tau_for_scores, 'SpearmanCorrelation': spearman_correlation_for_scores,
'ZeroOneRankLoss': zero_one_rank_loss_for_scores,
'ZeroOneRankLossTies': zero_one_rank_loss_for_scores_ties, "ZeroOneAccuracy": zero_one_accuracy_for_scores})
ranking_metrics_2 = OrderedDict(
{'KendallsTau': kendalls_mean_np, 'SpearmanCorrelation': spearman_mean_np,
'ZeroOneRankLoss': zero_one_rank_loss_for_scores_np,
'ZeroOneRankLossTies': zero_one_rank_loss_for_scores_ties_np, "ZeroOneAccuracy": zero_one_accuracy_np})
discrete_choice_metrics = OrderedDict(
{'categorical_accuracy': categorical_accuracy, 'CategoricalCrossEntropy': categorical_crossentropy})
lp_metric_dict = {
OBJECT_RANKING: ranking_metrics_2,
LABEL_RANKING: ranking_metrics,
DYAD_RANKING: ranking_metrics,
DISCRETE_CHOICE: discrete_choice_metrics
}
dataset_options_dict = {OBJECT_RANKING: (object_ranking_datasets)}
rankers_dict = {OBJECT_RANKING: object_rankers}
ERROR_OUTPUT_STRING = 'Out of sample error %s : %0.4f'
def get_ranker_and_dataset_functions(ranker_name, dataset_name, dataset_function_params, problem):
rankers = rankers_dict[problem]
datasets = dataset_options_dict[problem]
ranker = rankers[ranker_name]
dataset_reader = datasets[dataset_name]
dataset_reader = dataset_reader(**dataset_function_params)
return ranker, dataset_reader
def get_applicable_ranker_dataset(dataset_name, ranker_name, problem):
rankers = rankers_dict[problem]
datasets = dataset_options_dict[problem]
if ranker_name not in rankers:
ranker_name = BORDA_ZERO
if dataset_name not in datasets:
dataset_name = 'medoid'
return dataset_name, ranker_name
def get_ranker_parameters(ranker_name, n_features, n_objects, dataset_name, dataset_function_params, epochs=1000):
parameter_ranges = dict()
ranker_params = {'n_objects': n_objects, "n_object_features": n_features, "n_features": n_features}
if ranker_name == BORDA_ZERO or ranker_name == BORDA_RANKNET_ZERO:
ranker_params['add_zeroth_order_model'] = True
fit_params = {'epochs': epochs}
fit_params['log_callbacks'] = []
if dataset_name in [SENTENCE_ORDERING, HYPER_VOLUME, DEPTH, LETOR]:
fit_params['log_callbacks'] = [LRScheduler()]
parameter_ranges[BATCH_SIZE] = (512, 1024)
parameter_ranges[LEARNING_RATE] = (1e-5, 1e-4, LOG_UNIFORM)
ranker_params["optimizer"] = SGD(lr=1e-5, momentum=0.9, nesterov=True)
if ranker_name in [GENERAL_RANKER, BORDA_ZERO, BORDA, BORDA_RANKNET_ZERO, BORDA_RANKNET] and dataset_name not in [DEPTH,
HYPER_VOLUME, LETOR]:
ranker_params["loss_function"] = smooth_rank_loss
if dataset_name == LETOR:
parameter_ranges[LEARNING_RATE] = (1e-5, 1e-3, LOG_UNIFORM)
parameter_ranges[BATCH_SIZE] = (1024, 2048)
if dataset_name == DEPTH:
parameter_ranges[LEARNING_RATE] = (1e-5, 1e-3, LOG_UNIFORM)
parameter_ranges[BATCH_SIZE] = (256, 1024)
if ranker_name in [BORDA_ZERO, BORDA, BORDA_RANKNET_ZERO, BORDA_RANKNET, RANKNET, CMPNET]:
ranker_params["optimizer"] = "adam"
parameter_ranges[N_HIDDEN_SET_LAYERS] = parameter_ranges[N_HIDDEN_JOINT_LAYERS] = (5, 20)
parameter_ranges[N_HIDDEN_SET_UNITS] = parameter_ranges[N_HIDDEN_JOINT_UNITS] = (256, 1024)
if ranker_name in [BORDA_ZERO, BORDA, GENERAL_RANKER, RANKNET, CMPNET, BORDA_RANKNET_ZERO, BORDA_RANKNET]:
fit_params['log_callbacks'].append(DebugOutput())
ranker_params["use_early_stopping"] = True
if ranker_name == GENERAL_RANKER and dataset_name == SENTENCE_ORDERING and dataset_function_params.get(
"train_obj", 0) == 0:
fit_params = {'epochs': 35, "inner_epochs": 1, "min_bucket_size": 500, "validation_split": None}
return ranker_params, fit_params, parameter_ranges
def get_duration_microsecond(duration):
time = int(re.findall(r'\d+', duration)[0])
d = duration.split(str(time))[1].upper()
options = {"D": 24 * 60 * 60 * 1e6, "H": 60 * 60 * 1e6, "M": 60 * 1e6}
return options[d] * time
def get_value(v):
try:
x = int(v.strip())
except ValueError:
x = str(v.strip())
return x
def get_dataset_str(dataset_function_params, dataset_name):
dataset_function_params = dict(
(get_value(k), get_value(v)) for k, v in
(item.split(':') for item in dataset_function_params.split(',')))
keys = dataset_arg_options[dataset_name]
if keys is not None:
if isinstance(keys, list):
string = ''
for key in keys:
if dataset_function_params.get(key, 0) != 0:
string = string + "{}_{}".format(key, dataset_function_params[key])
dataset_str = "{}_{}".format(string, dataset_name)
else:
dataset_str = "{}_{}_{}".format(keys, str(dataset_function_params[keys]), dataset_name)
else:
dataset_str = dataset_name
return dataset_function_params, dataset_str
def log_test_train_data(X_train, X_test, logger):
if isinstance(X_train, dict) and isinstance(X_test, dict):
n_instances, n_objects, n_features = X_train[list(X_train.keys())[0]].shape
logger.info("instances {} objects {} features {}".format(n_instances, n_objects, n_features))
logger.info("Using Test Set dictionary of rankings with lengths {}".format(X_test.keys()))
logger.info("Using Training Set dictionary of rankings with lengths {}".format(X_train.keys()))
if not isinstance(X_test, dict):
n_i, n_o, n_f = X_test.shape
logger.info("Test Set instances {} objects {} features {}".format(n_i, n_o, n_f))
if not isinstance(X_train, dict):
n_instances, n_objects, n_features = X_train.shape
logger.info("Train Set instances {} objects {} features {}".format(n_instances, n_objects, n_features))
return n_features, n_objects
def get_optimizer(logger, optimizer_path, n_iter):
logger.info('Retrieving model stored at: {}'.format(optimizer_path))
try:
optimizer = load(optimizer_path)
logger.info('Loading model stored at: {}'.format(optimizer_path))
except KeyError:
logger.error('Cannot open the file {}'.format(optimizer_path))
optimizer = None
except ValueError:
logger.error('Cannot open the file {}'.format(optimizer_path))
optimizer = None
except FileNotFoundError:
logger.error('No such file or directory: {}'.format(optimizer_path))
optimizer = None
if optimizer is not None:
finished_iterations = np.array(optimizer.yi).shape[0]
if finished_iterations == 0:
optimizer = None
logger.info('Optimizer did not finish any iterations so setting optimizer to null')
else:
n_iter = n_iter - finished_iterations
if n_iter < 0:
n_iter = 0
logger.info('Iterations already done: {} and running iterations {}'.format(finished_iterations, n_iter))
return optimizer, n_iter