-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathdefault.py
290 lines (249 loc) · 10.4 KB
/
default.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
from typing import List, Optional, Union
from yacs.config import CfgNode as CN
from habitat_extensions.config.default import get_extended_config as get_task_config
# -----------------------------------------------------------------------------
# EXPERIMENT CONFIG
# -----------------------------------------------------------------------------
_C = CN()
_C.BASE_TASK_CONFIG_PATH = "habitat_extensions/config/vlnce_task.yaml"
_C.TASK_CONFIG = CN() # task_config will be stored as a config node
_C.CMD_TRAILING_OPTS = [] # store command line options as list of strings
_C.TRAINER_NAME = "robo_vln_trainer"
_C.ENV_NAME = "VLNCEDaggerEnv"
_C.SIMULATOR_GPU_ID = [0]
_C.TORCH_GPU_ID = 1
_C.NUM_PROCESSES = 4
_C.VIDEO_OPTION = [] # options: "disk", "tensorboard"
_C.VIDEO_DIR = "videos/debug"
_C.TENSORBOARD_DIR = "data/tensorboard_dirs/debug"
# _C.SENSORS = ["RGB_SENSOR", "DEPTH_SENSOR", "SEMANTIC_SENSOR"]
_C.SENSORS = ["RGB_SENSOR", "DEPTH_SENSOR"]
_C.CHECKPOINT_FOLDER = "data/checkpoints"
_C.LOG_FILE = "train.log"
_C.EVAL_CKPT_PATH_DIR = "data/checkpoints" # path to ckpt or path to ckpts dir
_C.PLOT_ATTENTION = True
# -----------------------------------------------------------------------------
# DDP CONFIG
# -----------------------------------------------------------------------------
_C.DDP = CN()
_C.DDP.world_size = 1
_C.DDP.dist_url = 'env://'
_C.DDP.rank = 0
_C.DDP.gpu = 0
_C.DDP.distributed = False
_C.DDP.dist_backend = 'nccl'
# -----------------------------------------------------------------------------
# EVAL CONFIG
# -----------------------------------------------------------------------------
_C.EVAL = CN()
# The split to evaluate on
_C.EVAL.SPLIT = "val_seen"
_C.EVAL.USE_CKPT_CONFIG = True
_C.EVAL.EPISODE_COUNT = 2
_C.EVAL.EVAL_NONLEARNING = False
_C.EVAL.NONLEARNING = CN()
_C.EVAL.NONLEARNING.AGENT = "RandomAgent"
_C.EVAL.VAL_LOG_DIR = "validation_logging"
# -----------------------------------------------------------------------------
# DAGGER ENVIRONMENT CONFIG
# -----------------------------------------------------------------------------
_C.DAGGER = CN()
_C.DAGGER.LR = 1e-4
_C.DAGGER.ITERATIONS = 1
_C.DAGGER.EPOCHS = 10
_C.DAGGER.UPDATE_SIZE = 5000
_C.DAGGER.BATCH_SIZE = 3
_C.DAGGER.P = 1.0
_C.DAGGER.time_step = 1.0 / (30)
_C.DAGGER.LMDB_MAP_SIZE = 2.7e12
_C.DAGGER.LMDB_EVAL_SIZE = 1e11
_C.DAGGER.tbptt_steps = 100
_C.DAGGER.split_dim = 0
_C.DAGGER.COLLECT_DATA_SPLIT = 'train'
_C.DAGGER.INTER_MODULE_ATTN = False
# How often to commit the writes to the DB, less commits is
# better, but everything must be in memory until a commit happens/
_C.DAGGER.LMDB_COMMIT_FREQUENCY = 500
_C.DAGGER.LMDB_STORE_FREQUENCY = 5
_C.DAGGER.USE_IW = True
# If True, load precomputed features directly from LMDB_FEATURES_DIR.
_C.DAGGER.PRELOAD_LMDB_FEATURES = False
_C.DAGGER.LMDB_FEATURES_DIR = "data/trajectories_dirs/debug/trajectories.lmdb"
_C.DAGGER.LMDB_EVAL_DIR = "data/trajectories_dirs/debug/trajectories.lmdb"
# load an already trained model for fine tuning
_C.DAGGER.LOAD_FROM_CKPT = False
_C.DAGGER.CKPT_TO_LOAD = "data/checkpoints/ckpt.0.pth"
# -----------------------------------------------------------------------------
# MODELING CONFIG
# -----------------------------------------------------------------------------
_C.MODEL = CN()
# on GT trajectories in the training set
_C.MODEL.inflection_weight_coef = 3.2
_C.MODEL.ablate_depth = False
_C.MODEL.ablate_rgb = False
_C.MODEL.ablate_instruction = False
_C.MODEL.ablate_sem_attn = False
_C.MODEL.INSTRUCTION_ENCODER = CN()
_C.MODEL.INSTRUCTION_ENCODER.num_layers = 1
_C.MODEL.INSTRUCTION_ENCODER.vocab_size = 2504
_C.MODEL.INSTRUCTION_ENCODER.max_length = 200
_C.MODEL.INSTRUCTION_ENCODER.use_pretrained_embeddings = True
_C.MODEL.INSTRUCTION_ENCODER.embedding_file = (
"data/datasets/robo_vln_v1/embeddings.json.gz"
)
_C.MODEL.INSTRUCTION_ENCODER.dataset_vocab = (
"data/datasets/R2R_VLNCE_v1_preprocessed/train/train.json.gz"
)
_C.MODEL.INSTRUCTION_ENCODER.fine_tune_embeddings = False
_C.MODEL.INSTRUCTION_ENCODER.embedding_size = 50
_C.MODEL.INSTRUCTION_ENCODER.hidden_size = 256
_C.MODEL.INSTRUCTION_ENCODER.rnn_type = "LSTM"
_C.MODEL.INSTRUCTION_ENCODER.final_state_only = True
_C.MODEL.INSTRUCTION_ENCODER.bidirectional = False
_C.MODEL.INSTRUCTION_ENCODER.dropout_ratio = 0.25
_C.MODEL.INSTRUCTION_ENCODER.is_bert = False
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER = CN()
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.N = 1
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.d_in = 768
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.d_model = 256
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.h = 4
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.d_ff = 1024
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.dropout = 0.2
_C.MODEL.TRANSFORMER_INSTRUCTION_ENCODER.is_bert = True
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER = CN()
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.N = 1
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.d_in = 512
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.d_out = 256
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.d_model = 256
# _C.MODEL.IMAGE_CROSS_MODAL_ENCODER.d_model = 512 for Hierarchical CMA without vlnce
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.h = 2
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.d_ff = 1024
_C.MODEL.IMAGE_CROSS_MODAL_ENCODER.dropout = 0.2
_C.MODEL.ACTION_DECODER_TRANFORMER = CN()
_C.MODEL.ACTION_DECODER_TRANFORMER.N = 1
_C.MODEL.ACTION_DECODER_TRANFORMER.in_features = 32
_C.MODEL.ACTION_DECODER_TRANFORMER.fc_output = 512
_C.MODEL.ACTION_DECODER_TRANFORMER.d_model = 512
_C.MODEL.ACTION_DECODER_TRANFORMER.h = 4
_C.MODEL.ACTION_DECODER_TRANFORMER.d_ff = 1024
_C.MODEL.ACTION_DECODER_TRANFORMER.dropout = 0.1
_C.MODEL.INTER_MODULE_ATTN = CN()
_C.MODEL.INTER_MODULE_ATTN.N = 1
_C.MODEL.INTER_MODULE_ATTN.in_features = 512
_C.MODEL.INTER_MODULE_ATTN.fc_output = 512
_C.MODEL.INTER_MODULE_ATTN.d_model = 512
_C.MODEL.INTER_MODULE_ATTN.h = 4
_C.MODEL.INTER_MODULE_ATTN.d_ff = 1024
_C.MODEL.INTER_MODULE_ATTN.dropout = 0.1
_C.MODEL.VISUAL_LING_ATTN = CN()
_C.MODEL.VISUAL_LING_ATTN.N = 1
_C.MODEL.VISUAL_LING_ATTN.vis_in_features = 256
_C.MODEL.VISUAL_LING_ATTN.ins_in_features = 768
_C.MODEL.VISUAL_LING_ATTN.fc_output = 512
_C.MODEL.VISUAL_LING_ATTN.d_model = 256
_C.MODEL.VISUAL_LING_ATTN.h = 4
_C.MODEL.VISUAL_LING_ATTN.d_ff = 1024
_C.MODEL.VISUAL_LING_ATTN.dropout = 0.25
_C.MODEL.SEM_MAP_TRANSFORMER = CN()
_C.MODEL.SEM_MAP_TRANSFORMER.embedding_dim = 128
_C.MODEL.SEM_MAP_TRANSFORMER.layer_norm_eps=1e-12
_C.MODEL.SEM_MAP_TRANSFORMER.N = 1
_C.MODEL.SEM_MAP_TRANSFORMER.d_in = 128
_C.MODEL.SEM_MAP_TRANSFORMER.d_model = 512
_C.MODEL.SEM_MAP_TRANSFORMER.d_out = 256
_C.MODEL.SEM_MAP_TRANSFORMER.h = 4
_C.MODEL.SEM_MAP_TRANSFORMER.d_ff = 1024
_C.MODEL.SEM_MAP_TRANSFORMER.dropout = 0.1
_C.MODEL.SEM_MAP_TRANSFORMER.downsample_size = 20
_C.MODEL.SEM_MAP_TRANSFORMER.n_output = 512
_C.MODEL.RGB_ENCODER = CN()
# 'SimpleRGBCNN' or 'TorchVisionResNet50'
_C.MODEL.RGB_ENCODER.cnn_type = "TorchVisionResNet50"
_C.MODEL.RGB_ENCODER.output_size = 256 # for CMA + Seq2Seq + InterModule Attention + hierarchical CMA VLNCE
#_C.MODEL.RGB_ENCODER.output_size = 512 # Base Hierarchical + Cross Modal Hierarchical + Flat Aux
_C.MODEL.RGB_ENCODER.resnet_output_size = 256
_C.MODEL.DEPTH_ENCODER = CN()
# 'VlnResnetDepthEncoder' or 'SimpleDepthCNN'
_C.MODEL.DEPTH_ENCODER.cnn_type = "VlnResnetDepthEncoder"
_C.MODEL.DEPTH_ENCODER.output_size = 128
# type of resnet to use
_C.MODEL.DEPTH_ENCODER.backbone = "resnet50"
# path to DDPPO resnet weights
_C.MODEL.DEPTH_ENCODER.ddppo_checkpoint = "data/ddppo-models/gibson-2plus-resnet50.pth"
_C.MODEL.STATE_ENCODER = CN()
_C.MODEL.STATE_ENCODER.hidden_size = 512 # for Inter Module Attention + Cross Attention + Base Hierarchical
#_C.MODEL.STATE_ENCODER.hidden_size = 256 #
_C.MODEL.STATE_ENCODER.rnn_type = "LSTM"
_C.MODEL.SEQ2SEQ = CN()
_C.MODEL.SEQ2SEQ.use_prev_action = False
_C.MODEL.HIERARCHICAL = CN()
_C.MODEL.SEQ2SEQ.use_prev_action = False
_C.MODEL.CMA = CN()
_C.MODEL.CMA.use = False
# Use the state encoding model in RCM. If false,
# will just concat inputs and run an RNN over them
_C.MODEL.CMA.rcm_state_encoder = False
_C.MODEL.CMA.use_prev_action = False
_C.MODEL.PROGRESS_MONITOR = CN()
_C.MODEL.PROGRESS_MONITOR.use = False
_C.MODEL.PROGRESS_MONITOR.alpha = 1.0 # loss multiplier
_C.MODEL.SEM_ATTN_ENCODER = CN()
_C.MODEL.SEM_ATTN_ENCODER.use = False
_C.MODEL.SEM_ATTN_ENCODER.hidden_size = 256
_C.MODEL.LANG_ATTN= CN()
_C.MODEL.LANG_ATTN.use = False
_C.MODEL.LANG_ATTN.hidden_size = 256
_C.MODEL.FLAT_AUX_LOSS= CN()
_C.MODEL.FLAT_AUX_LOSS.use = False
_C.MODEL.SEM_TEXT_ATTN = CN()
_C.MODEL.SEM_TEXT_ATTN.use = False
_C.MODEL.SEM_TEXT_ATTN.hidden_size = 256
_C.MODEL.TRANSFORMER = CN()
_C.MODEL.TRANSFORMER.use = False
_C.MODEL.TRANSFORMER.output_size = 512
_C.MODEL.TRANSFORMER.use_prev_action = True
_C.MODEL.TRANSFORMER.lr = 1e-4
_C.MODEL.TRANSFORMER.scheduler_patience = 1e-4
_C.MODEL.TRANSFORMER.weight_decay = 1e-3
_C.MODEL.TRANSFORMER.lr_drop = 4
_C.MODEL.TRANSFORMER.split_gpus = False
_C.MODEL.TRANSFORMER.hidden_size = 512
_C.MODEL.HYBRID_STATE_DECODER = CN()
_C.MODEL.HYBRID_STATE_DECODER.in_features = 512
_C.MODEL.HYBRID_STATE_DECODER.hidden_size = 512
_C.MODEL.HYBRID_STATE_DECODER.fc_output = 512
_C.MODEL.HYBRID_STATE_DECODER.rnn_type = "LSTM"
_C.MODEL.HYBRID_STATE_DECODER.N = 1
_C.MODEL.HYBRID_STATE_DECODER.d_in = 512
_C.MODEL.HYBRID_STATE_DECODER.d_out = 256
_C.MODEL.HYBRID_STATE_DECODER.d_model = 512
_C.MODEL.HYBRID_STATE_DECODER.h = 4
_C.MODEL.HYBRID_STATE_DECODER.d_ff = 1024
_C.MODEL.HYBRID_STATE_DECODER.dropout = 0.1
_C.MODEL.HYBRID_STATE_DECODER.RNN_output_size = 512
_C.MODEL.HYBRID_STATE_DECODER.prev_action_embedding_dim = 32
def get_config(
config_paths: Optional[Union[List[str], str]] = None, opts: Optional[list] = None
) -> CN:
r"""Create a unified config with default values overwritten by values from
`config_paths` and overwritten by options from `opts`.
Args:
config_paths: List of config paths or string that contains comma
separated list of config paths.
opts: Config options (keys, values) in a list (e.g., passed from
command line into the config. For example, `opts = ['FOO.BAR',
0.5]`. Argument can be used for parameter sweeping or quick tests.
"""
config = _C.clone()
if config_paths:
if isinstance(config_paths, str):
config_paths = [config_paths]
for config_path in config_paths:
config.merge_from_file(config_path)
if config.BASE_TASK_CONFIG_PATH != "":
config.TASK_CONFIG = get_task_config(config.BASE_TASK_CONFIG_PATH)
if opts:
config.CMD_TRAILING_OPTS = opts
config.merge_from_list(opts)
config.freeze()
return config