/
generate_eval_config.py
311 lines (278 loc) · 11.3 KB
/
generate_eval_config.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
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
import os
from copy import deepcopy
import numpy as np
import pandas as pd
from rlq_scheduler.common.config_helper import MultiRunConfigHelper, GlobalConfigHelper
from rlq_scheduler.common.object_handler import create_object_handler, MinioObjectHandler
from rlq_scheduler.common.plot_utils.stats import get_stats_dataframe, load_agents_data, print_status
from rlq_scheduler.common.stats import RunStats
from rlq_scheduler.common.trajectory_saver.database import Database
from rlq_scheduler.common.utils.filesystem import save_file, ROOT_DIR
from rlq_scheduler.common.utils.logger import get_logger
RESULT_FOLDER = 'waiting-time'
RESULT_FOLDER_OUTPUT = 'eval-new-waiting-time'
OUTPUT_FILES = 'config/kube/deployments/waiting-time/new'
MINIO_BUCKET = 'gtraces1'
AGENT_NAME = 'DoubleDQN'
AGENT_CODE = 'double-dqn'
AGENT_PARAM = ('lr', 'layers')
LOAD_SEEDS_FROM_DB = None # or None
DB_NAME = 'sb_waiting_time'
DB_COLLECTION = 'waiting-time'
PARAM_NAME_MAPPING = {
'delta': 'delta',
'lr': 'learning_rate',
'layers': 'network_config.parameters.hidden_layers'
}
PARAM_NAME_TYPE_MAPPING = {
'delta': float,
'lr': float,
'layers': int
}
TASK_GENERATOR_SEED = 200
N_TASKS = 13730
N_BOOTSTRAP_TASKS = 0
REWARD_FUNCTION = 'waiting-time'
REWARD_FUNCTION_PARAMETERS = {}
LOAD_MODEL = False
TRAIN = True
ADD_BASELINES = False
AGENT_PARAMETERS = {
'LinUCB': {
'type': AGENT_CODE,
'delta': 2,
'parameters': [
{
'type': 'agent_global',
'name': 'path',
'param': 'load_model_config',
'mode': 'array',
'seed': None,
'values': []
}
]
},
'DoubleDQN': {
'type': AGENT_CODE,
'experience_replay_capacity': 4000,
'learning_rate': 0.001,
'gamma': 0.99,
'batch_size': 64,
'target_net_update_frequency': 400,
'epsilon': {
'type': 'linear-decay',
'parameters': {'start': 0.65, 'end': 0.1, 'total': 6000}
},
'optimizer': 'adam',
'network_config': {
'type': 'fully-connected',
'parameters': {'hidden_layers': 3}
},
'parameters': [
{
'type': 'agent_global',
'name': 'path',
'param': 'load_model_config',
'mode': 'array',
'seed': None,
'values': []
}
]
}
}
agents = {
AGENT_NAME: {
'runs_names': [],
'runs_stats': []
}}
BASELINE_PARAMETERS = {
'type': 'agent_global',
'name': 'random_seed',
'mode': 'array',
'seed': None,
'values': []
}
test_runs = [
{'seed': 3,
'run_code': '6c47d1b6-fceb-44d2-9c59-3ead3f745dc9',
'total_reward': -9924.397213935852},
{'seed': 4,
'run_code': 'e723d898-a9b4-463d-89f0-5ccbe7322733',
'total_reward': -26692.695888996124},
{'seed': 5,
'run_code': '38b287b7-338a-4d5d-a2a9-a2b7f39a969e',
'total_reward': -20220.14770269394},
{'seed': 6,
'run_code': '91002ee0-57e1-4b12-9729-79cf95643668',
'total_reward': -16641.10775566101},
{'seed': 7,
'run_code': 'e7e2a640-36ce-497c-bb91-1aeb2ceddaa3',
'total_reward': -61731.80339550972}
]
test_param = (2.0, 3)
def set_best_params(agent_config, parameters):
if isinstance(parameters, float) or isinstance(parameters, np.float64):
# single param
p_name = AGENT_PARAM
p_type = PARAM_NAME_TYPE_MAPPING[p_name]
p_value = p_type(parameters)
set_param_value(agent_config, p_value, p_name)
elif isinstance(parameters, tuple):
for i, p in enumerate(parameters):
p_name = AGENT_PARAM[i]
p_type = PARAM_NAME_TYPE_MAPPING[p_name]
p_value = p_type(p)
set_param_value(agent_config, p_value, p_name)
def set_param_value(agent_config, p_value, p_name):
mapping = PARAM_NAME_MAPPING[p_name]
if '.' not in mapping:
agent_config[mapping] = p_value
else:
parts = mapping.split('.')
nested = None
for p in parts[0:-1]:
if nested is None:
nested = agent_config[p]
else:
nested = nested[p]
nested[parts[-1]] = p_value
def param_to_array(parameters):
if isinstance(parameters, float) or isinstance(parameters, np.float64) or isinstance(parameters, str):
# single param
return [parameters]
elif isinstance(parameters, tuple):
return list(parameters)
def filter_df(df, agent_name, best_p):
if agent_name == 'LinUCB':
return df[df.delta == best_p]
elif agent_name == 'DoubleDQN':
return df[(df.lr == best_p[0]) & (df.layers == best_p[1])]
def load_agents_data_from_run_seeds_and_db(
agents_data: dict,
object_handle: MinioObjectHandler,
result_folder: str,
database: Database,
db_name: str,
collection: str,
seeds: list,
mode='train'
):
for agent_name, agent_info in agents_data.items():
print(f'Loading runs name for agent {agent_name}')
runs_codes = [r['run_code'] for r in database.client[db_name][collection].find({
'agent_parameters.agent_seed': {'$in': seeds}, 'agent_type': agent_name})]
paths_to_load = []
full_path = os.path.join(result_folder, "results", mode, agent_name)
folder_files = handler.list_objects_name(f'{full_path}/')
for path in folder_files:
code = path.split('/')[-1].split('_')[0]
if code in runs_codes:
paths_to_load.append(path)
agent_info['runs_names'] = paths_to_load
print(f'Loading {len(agent_info["runs_names"])} runs for agent {agent_name}')
for i, run_path in enumerate(agent_info['runs_names']):
agent_info['runs_stats'].append(RunStats.from_dict(object_handle.load(run_path)))
print_status(i + 1, len(agent_info['runs_names']), f'Loading agent runs')
print(f'\nLoaded all the {len(agent_info["runs_names"])} runs for agent {agent_name}')
return agents_data
def load_agents_data_efficient(
agents_data: dict,
o_handler: MinioObjectHandler,
result_folder: str,
mode='train'
):
dfs = []
for agent_name, agent_info in agents_data.items():
print(f'Loading runs name for agent {agent_name}')
result_path = os.path.join(result_folder, "results", mode, agent_name)
agent_info['runs_names'] = o_handler.list_objects_name(f'{result_path}/', recursive=False)
print(f'Loading {len(agent_info["runs_names"])} runs for agent {agent_name}')
for i, run_path in enumerate(agent_info['runs_names']):
stats = RunStats.from_dict(o_handler.load(run_path))
agent_info['runs_stats'] = [stats]
dfs.append(get_stats_dataframe(
agents,
{
'label': 'total_reward',
'name': 'reward',
'aggregation': 'sum',
'skip': 2999 if AGENT_CODE == 'double-dqn' else 0
}
))
print_status(i + 1, len(agent_info['runs_names']), f'Loading agent runs')
print(f'\nLoaded all the {len(agent_info["runs_names"])} runs for agent {agent_name}')
return pd.concat(dfs, ignore_index=True)
if __name__ == '__main__':
global_config = GlobalConfigHelper(config_path='config/global.yml')
global_config.config['object_handler']['default_bucket'] = MINIO_BUCKET
logger = get_logger(global_config.logger())
handler = create_object_handler(global_config, logger)
db = Database(global_config, logger)
logger.info('Starting results loading')
if LOAD_SEEDS_FROM_DB is not None:
agents = load_agents_data_from_run_seeds_and_db(
agents, handler, RESULT_FOLDER, db, DB_NAME, DB_COLLECTION, LOAD_SEEDS_FROM_DB)
rewards_df = get_stats_dataframe(
agents,
{
'label': 'total_reward',
'name': 'reward',
'aggregation': 'sum',
'skip': 2999 if AGENT_CODE == 'double-dqn' else 0
}
)
else:
rewards_df = load_agents_data_efficient(agents, handler, RESULT_FOLDER)
agent_df = rewards_df[rewards_df.agent == AGENT_NAME]
agent_no_multi_df = agent_df[agent_df.reward_multiplier == 1]
print(f'Best {AGENT_NAME} without reward multiplier param and best param run codes\n')
best_param = agent_no_multi_df.groupby(param_to_array(AGENT_PARAM)).total_reward.mean().idxmax()
best_value = agent_no_multi_df.groupby(param_to_array(AGENT_PARAM)).total_reward.mean().max()
print('best param: \t\t{}\t\t\t\t\t\t max mean: \t{}'.format(best_param, best_value))
print('------------------------')
best_runs = []
filtered_df = filter_df(agent_no_multi_df, AGENT_NAME, best_param)
for _, value in filtered_df.iterrows():
best_runs.append({'seed': value.seed, 'run_code': value.run_code, 'total_reward': value.total_reward})
best_runs.sort(key=lambda x: x['seed'])
for value in best_runs:
print(f'seed {value["seed"]} run code: \t{value["run_code"]}\t\t reward: \t{value["total_reward"]}')
template = MultiRunConfigHelper(config_path='config/multi_run_config.yml')
config = deepcopy(template.config)
config['global']['seeds']['auto'] = False
config['global']['seeds']['n_runs'] = 1
config['global']['seeds']['task_generator'] = TASK_GENERATOR_SEED
config['global']['seeds']['agents'] = [3]
config['task_generator']['bootstrapping']['tasks_to_generate'] = N_BOOTSTRAP_TASKS
config['task_generator']['tasks_to_generate'][0]['tasks_to_generate'] = N_TASKS
config['save_properties']['run_name_prefix'] = RESULT_FOLDER_OUTPUT
config['functions']['reward_function']['type'] = REWARD_FUNCTION
config['functions']['reward_function']['extra_parameters'] = REWARD_FUNCTION_PARAMETERS
config['global_agent_config']['load_model_config']['load'] = LOAD_MODEL
config['global_agent_config']['train'] = TRAIN
agents_config = []
for data in best_runs:
main_agent_config = deepcopy(AGENT_PARAMETERS[AGENT_NAME])
set_best_params(main_agent_config, best_param)
model_path = ''
files = handler.list_objects_name(f'{RESULT_FOLDER}/models/train/{AGENT_NAME}/{data["run_code"]}')
main_agent_config['parameters'][0]['values'].append(files[0])
main_agent_config['parameters'][0]['seed'] = data['seed']
agents_config.append(main_agent_config)
if ADD_BASELINES:
random_parameters = deepcopy(BASELINE_PARAMETERS)
random_parameters['seed'] = data['seed']
random_parameters['values'].append(data['seed'])
random = {'type': 'random', 'parameters': [random_parameters]}
lru_parameters = deepcopy(BASELINE_PARAMETERS)
lru_parameters['seed'] = data['seed']
lru_parameters['values'].append(data['seed'])
lru = {'type': 'lru', 'parameters': [lru_parameters]}
agents_config.append(random)
agents_config.append(lru)
config['agents'] = agents_config
save_file(os.path.join(ROOT_DIR, OUTPUT_FILES),
f'{AGENT_CODE}_eval-config.yml',
config,
is_yml=True)
logger.info(f'Created all the configuration files and saved in {OUTPUT_FILES}')