Comments (20)
erikwijmans
commented on September 12, 2024
1
You can change resnet50 to resnet18 in the config, that will improve training speed.
from habitat-challenge.
commented on September 12, 2024
1
@erikwijmans As you suggested I trained DD-PPO model with resnet18 backbone.
When I tried to evaluate it, I got the following error:
Traceback (most recent call last):
File "agent.py", line 165, in <module>
main()
File "agent.py", line 155, in main
agent = DDPPOAgent(config)
File "agent.py", line 88, in __init__
for k, v in ckpt["state_dict"].items()
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 847, in load_state_dict
self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for PointNavResNetPolicy:
Missing key(s) in state_dict: "net.visual_encoder.backbone.layer1.0.convs.6.weight", "net.visual_encoder.backbone.layer1.0.convs.7.weight", "net.visual_encoder.backbone.layer1.0.convs.7.bias", "net.visual_encoder.backbone.layer1.0.downsample.0.weight", "net.visual_encoder.backbone.layer1.0.downsample.1.weight", "net.visual_encoder.backbone.layer1.0.downsample.1.bias", "net.visual_encoder.backbone.layer1.1.convs.6.weight", "net.visual_encoder.backbone.layer1.1.convs.7.weight", "net.visual_encoder.backbone.layer1.1.convs.7.bias", "net.visual_encoder.backbone.layer1.2.convs.0.weight", "net.visual_encoder.backbone.layer1.2.convs.1.weight", "net.visual_encoder.backbone.layer1.2.convs.1.bias", "net.visual_encoder.backbone.layer1.2.convs.3.weight", "net.visual_encoder.backbone.layer1.2.convs.4.weight", "net.visual_encoder.backbone.layer1.2.convs.4.bias", "net.visual_encoder.backbone.layer1.2.convs.6.weight", "net.visual_encoder.backbone.layer1.2.convs.7.weight", "net.visual_encoder.backbone.layer1.2.convs.7.bias", "net.visual_encoder.backbone.layer2.0.convs.6.weight", "net.visual_encoder.backbone.layer2.0.convs.7.weight", "net.visual_encoder.backbone.layer2.0.convs.7.bias", "net.visual_encoder.backbone.layer2.1.convs.6.weight", "net.visual_encoder.backbone.layer2.1.convs.7.weight", "net.visual_encoder.backbone.layer2.1.convs.7.bias", "net.visual_encoder.backbone.layer2.2.convs.0.weight", "net.visual_encoder.backbone.layer2.2.convs.1.weight", "net.visual_encoder.backbone.layer2.2.convs.1.bias", "net.visual_encoder.backbone.layer2.2.convs.3.weight", "net.visual_encoder.backbone.layer2.2.convs.4.weight", "net.visual_encoder.backbone.layer2.2.convs.4.bias", "net.visual_encoder.backbone.layer2.2.convs.6.weight", "net.visual_encoder.backbone.layer2.2.convs.7.weight", "net.visual_encoder.backbone.layer2.2.convs.7.bias", "net.visual_encoder.backbone.layer2.3.convs.0.weight", "net.visual_encoder.backbone.layer2.3.convs.1.weight", "net.visual_encoder.backbone.layer2.3.convs.1.bias", "net.visual_encoder.backbone.layer2.3.convs.3.weight", "net.visual_encoder.backbone.layer2.3.convs.4.weight", "net.visual_encoder.backbone.layer2.3.convs.4.bias", "net.visual_encoder.backbone.layer2.3.convs.6.weight", "net.visual_encoder.backbone.layer2.3.convs.7.weight", "net.visual_encoder.backbone.layer2.3.convs.7.bias", "net.visual_encoder.backbone.layer3.0.convs.6.weight", "net.visual_encoder.backbone.layer3.0.convs.7.weight", "net.visual_encoder.backbone.layer3.0.convs.7.bias", "net.visual_encoder.backbone.layer3.1.convs.6.weight", "net.visual_encoder.backbone.layer3.1.convs.7.weight", "net.visual_encoder.backbone.layer3.1.convs.7.bias", "net.visual_encoder.backbone.layer3.2.convs.0.weight", "net.visual_encoder.backbone.layer3.2.convs.1.weight", "net.visual_encoder.backbone.layer3.2.convs.1.bias", "net.visual_encoder.backbone.layer3.2.convs.3.weight", "net.visual_encoder.backbone.layer3.2.convs.4.weight", "net.visual_encoder.backbone.layer3.2.convs.4.bias", "net.visual_encoder.backbone.layer3.2.convs.6.weight", "net.visual_encoder.backbone.layer3.2.convs.7.weight", "net.visual_encoder.backbone.layer3.2.convs.7.bias", "net.visual_encoder.backbone.layer3.3.convs.0.weight", "net.visual_encoder.backbone.layer3.3.convs.1.weight", "net.visual_encoder.backbone.layer3.3.convs.1.bias", "net.visual_encoder.backbone.layer3.3.convs.3.weight", "net.visual_encoder.backbone.layer3.3.convs.4.weight", "net.visual_encoder.backbone.layer3.3.convs.4.bias", "net.visual_encoder.backbone.layer3.3.convs.6.weight", "net.visual_encoder.backbone.layer3.3.convs.7.weight", "net.visual_encoder.backbone.layer3.3.convs.7.bias", "net.visual_encoder.backbone.layer3.4.convs.0.weight", "net.visual_encoder.backbone.layer3.4.convs.1.weight", "net.visual_encoder.backbone.layer3.4.convs.1.bias", "net.visual_encoder.backbone.layer3.4.convs.3.weight", "net.visual_encoder.backbone.layer3.4.convs.4.weight", "net.visual_encoder.backbone.layer3.4.convs.4.bias", "net.visual_encoder.backbone.layer3.4.convs.6.weight", "net.visual_encoder.backbone.layer3.4.convs.7.weight", "net.visual_encoder.backbone.layer3.4.convs.7.bias", "net.visual_encoder.backbone.layer3.5.convs.0.weight", "net.visual_encoder.backbone.layer3.5.convs.1.weight", "net.visual_encoder.backbone.layer3.5.convs.1.bias", "net.visual_encoder.backbone.layer3.5.convs.3.weight", "net.visual_encoder.backbone.layer3.5.convs.4.weight", "net.visual_encoder.backbone.layer3.5.convs.4.bias", "net.visual_encoder.backbone.layer3.5.convs.6.weight", "net.visual_encoder.backbone.layer3.5.convs.7.weight", "net.visual_encoder.backbone.layer3.5.convs.7.bias", "net.visual_encoder.backbone.layer4.0.convs.6.weight", "net.visual_encoder.backbone.layer4.0.convs.7.weight", "net.visual_encoder.backbone.layer4.0.convs.7.bias", "net.visual_encoder.backbone.layer4.1.convs.6.weight", "net.visual_encoder.backbone.layer4.1.convs.7.weight", "net.visual_encoder.backbone.layer4.1.convs.7.bias", "net.visual_encoder.backbone.layer4.2.convs.0.weight", "net.visual_encoder.backbone.layer4.2.convs.1.weight", "net.visual_encoder.backbone.layer4.2.convs.1.bias", "net.visual_encoder.backbone.layer4.2.convs.3.weight", "net.visual_encoder.backbone.layer4.2.convs.4.weight", "net.visual_encoder.backbone.layer4.2.convs.4.bias", "net.visual_encoder.backbone.layer4.2.convs.6.weight", "net.visual_encoder.backbone.layer4.2.convs.7.weight", "net.visual_encoder.backbone.layer4.2.convs.7.bias".
size mismatch for net.visual_encoder.backbone.layer1.0.convs.0.weight: copying a param with shape torch.Size([32, 32, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 32, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer1.1.convs.0.weight: copying a param with shape torch.Size([32, 32, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 128, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer2.0.convs.0.weight: copying a param with shape torch.Size([64, 32, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 128, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer2.0.downsample.0.weight: copying a param with shape torch.Size([64, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([256, 128, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer2.0.downsample.1.weight: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]).
size mismatch for net.visual_encoder.backbone.layer2.0.downsample.1.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]).
size mismatch for net.visual_encoder.backbone.layer2.1.convs.0.weight: copying a param with shape torch.Size([64, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 256, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer3.0.convs.0.weight: copying a param with shape torch.Size([128, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 256, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer3.0.downsample.0.weight: copying a param with shape torch.Size([128, 64, 1, 1]) from checkpoint, the shape in current model is torch.Size([512, 256, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer3.0.downsample.1.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]).
size mismatch for net.visual_encoder.backbone.layer3.0.downsample.1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]).
size mismatch for net.visual_encoder.backbone.layer3.1.convs.0.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 512, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer4.0.convs.0.weight: copying a param with shape torch.Size([256, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 512, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer4.0.downsample.0.weight: copying a param with shape torch.Size([256, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([1024, 512, 1, 1]).
size mismatch for net.visual_encoder.backbone.layer4.0.downsample.1.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]).
size mismatch for net.visual_encoder.backbone.layer4.0.downsample.1.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]).
size mismatch for net.visual_encoder.backbone.layer4.1.convs.0.weight: copying a param with shape torch.Size([256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 1024, 1, 1]).
size mismatch for net.visual_encoder.compression.0.weight: copying a param with shape torch.Size([128, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 1024, 3, 3]).
Looks like it loads config for model with resnet50 backbone, but here is my config file:
TRAINER_NAME: "ddppo"
ENV_NAME: "NavRLEnv"
SIMULATOR_GPU_ID: 0
TORCH_GPU_ID: 0
VIDEO_OPTION: []
TENSORBOARD_DIR: "tb"
VIDEO_DIR: "video_dir"
TEST_EPISODE_COUNT: -1
EVAL_CKPT_PATH_DIR: "data/new_checkpoints"
NUM_PROCESSES: 8
SENSORS: ["RGB_SENSOR" , "DEPTH_SENSOR"]
CHECKPOINT_FOLDER: "data/new_checkpoints"
NUM_UPDATES: 1000000
LOG_INTERVAL: 10
CHECKPOINT_INTERVAL: 50
RL:
SLACK_REWARD: -0.001
SUCCESS_REWARD: 2.5
PPO:
# ppo params
clip_param: 0.2
ppo_epoch: 2
num_mini_batch: 2
value_loss_coef: 0.5
entropy_coef: 0.01
lr: 2.5e-4
eps: 1e-5
max_grad_norm: 0.2
num_steps: 64
use_gae: True
gamma: 0.99
tau: 0.95
use_linear_clip_decay: False
use_linear_lr_decay: False
reward_window_size: 50
use_normalized_advantage: False
hidden_size: 512
DDPPO:
sync_frac: 0.6
# The PyTorch distributed backend to use
distrib_backend: GLOO
# Visual encoder backbone
pretrained_weights: data/ddppo-models/gibson-2plus-resnet50.pth
# Initialize with pretrained weights
pretrained: False
# Initialize just the visual encoder backbone with pretrained weights
pretrained_encoder: False
# Whether or not the visual encoder backbone will be trained.
train_encoder: True
# Whether or not to reset the critic linear layer
reset_critic: True
# Model parameters
backbone: resnet18
rnn_type: LSTM
num_recurrent_layers: 2
I am wondering where could be the problem.
from habitat-challenge.
erikwijmans
commented on September 12, 2024
DATA_PATH: data/datasets/pointnav/gibson/v1/{split}/{split}.json.gz
This needs to be v2 (from here: https://github.com/facebookresearch/habitat-api#task-datasets). We changed the agent size for the challenge this year and that requires new episodes
from habitat-challenge.
commented on September 12, 2024
@erikwijmans thanks for fast answer!
I downloaded and changed the data to version 2.
DATASET:
TYPE: PointNav-v1
SPLIT: train
DATA_PATH: data/datasets/pointnav/gibson/v2/{split}/{split}.json.gz
I trained PPO model on new data and than wanted to evaluate it with docker, but got the following error. Looks like something is wrong with resolution, but I can't find out the specific issue.
Traceback (most recent call last):
File "ppo_agent.py", line 41, in <module>
main()
File "ppo_agent.py", line 30, in main
agent = PPOAgent(agent_config)
File "/habitat-api/habitat_baselines/agents/ppo_agents.py", line 92, in __init__
for k, v in ckpt["state_dict"].items()
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 830, in load_state_dict
self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for PointNavBaselinePolicy:
size mismatch for net.visual_encoder.cnn.6.weight: copying a param with shape torch.Size([512, 99712]) from checkpoint, the shape in current model is torch.Size([512, 184832]).
Here is my ppo_agent.py file:
import argparse
import habitat
import random
import numpy
import os
from habitat.config import Config
from habitat.config.default import get_config
from habitat_baselines.agents.ppo_agents import PPOAgent
def get_default_config():
c = Config()
c.INPUT_TYPE = "rgbd" #["blind", "rgb", "depth", "rgbd"]
c.MODEL_PATH = "models/ckpt.3.pth"
c.RESOLUTION = 640
c.HIDDEN_SIZE = 512
c.RANDOM_SEED = 7
c.PTH_GPU_ID = 1
c.GOAL_SENSOR_UUID = "pointgoal"
return c
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--evaluation", type=str, required=True, choices=["local", "remote"])
args = parser.parse_args()
agent_config = get_default_config()
agent = PPOAgent(agent_config)
if args.evaluation == "local":
challenge = habitat.Challenge(eval_remote=False)
else:
challenge = habitat.Challenge(eval_remote=True)
challenge.submit(agent)
if __name__ == "__main__":
main()
from habitat-challenge.
commented on September 12, 2024
@erikwijmans @dhruvbatra maybe you have some suggestions to the above error (with mismatched shapes) ?
Also, when I want to test the same ppo model using python -u habitat_baselines/run.py --exp-config habitat_baselines/config/pointnav/ppo_pointnav.yaml --run-type eval command I get the following error:
---
The active scene does not contain semantic annotations.
---
I0402 11:34:42.003789 2415 simulator.py:143] Loaded navmesh data/scene_datasets/gibson/Ribera.navmesh
I0402 11:34:42.004161 2415 simulator.py:155] Recomputing navmesh for agent's height 0.88 and radius 0.18.
I0402 11:34:42.011056 2415 PathFinder.cpp:338] Building navmesh with 127x175 cells
I0402 11:34:42.062223 2415 PathFinder.cpp:606] Created navmesh with 96 vertices 46 polygons
I0402 11:34:42.062256 2415 Simulator.cpp:403] reconstruct navmesh successful
2020-04-02 11:34:42,083 Initializing task Nav-v0
0%| | 1/994 [00:00<12:20, 1.34it/s]Traceback (most recent call last):
File "habitat_baselines/run.py", line 70, in <module>
main()
File "habitat_baselines/run.py", line 40, in main
run_exp(**vars(args))
File "habitat_baselines/run.py", line 66, in run_exp
trainer.eval()
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/common/base_trainer.py", line 108, in eval
checkpoint_index=prev_ckpt_ind,
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/rl/ppo/ppo_trainer.py", line 574, in _eval_checkpoint
metric_name=self.metric_uuid,
AttributeError: 'PPOTrainer' object has no attribute 'metric_uuid'
0%| | 1/994 [00:01<18:36, 1.12s/it]
Exception ignored in: <bound method VectorEnv.__del__ of <habitat.core.vector_env.VectorEnv object at 0x7f496cd2b160>>
Traceback (most recent call last):
File "/home/pryhoda/HabitatProject/habitat-api/habitat/core/vector_env.py", line 468, in __del__
self.close()
File "/home/pryhoda/HabitatProject/habitat-api/habitat/core/vector_env.py", line 350, in close
write_fn((CLOSE_COMMAND, None))
File "/home/pryhoda/anaconda3/envs/habitat/lib/python3.6/multiprocessing/connection.py", line 206, in send
self._send_bytes(_ForkingPickler.dumps(obj))
File "/home/pryhoda/anaconda3/envs/habitat/lib/python3.6/multiprocessing/connection.py", line 404, in _send_bytes
self._send(header + buf)
File "/home/pryhoda/anaconda3/envs/habitat/lib/python3.6/multiprocessing/connection.py", line 368, in _send
n = write(self._handle, buf)
BrokenPipeError: [Errno 32] Broken pipe
from habitat-challenge.
erikwijmans
commented on September 12, 2024
Not sure about the size mismatch. The AttributeError: 'PPOTrainer' object has no attribute 'metric_uuid' is fixed in facebookresearch/habitat-lab#357
from habitat-challenge.
commented on September 12, 2024
@erikwijmans confirm that with latest commit habitata-api:fix-video-gen the issue AttributeError: 'PPOTrainer' object has no attribute 'metric_uuid' has been solved.
from habitat-challenge.
commented on September 12, 2024
@erikwijmans coming back to previous issue with size mismatch. I have an assumption that the problem is in get_default_config() function for PPO agent. Specifically in c.RESOLUTION variable.
In 2019 challenge config the resolution was 256x256, and in 2020 challenge it is 640x480. But how properly set the new size up in configuration if there is only one variable c.RESOLUTION ?
from habitat-challenge.
erikwijmans
commented on September 12, 2024
CC @mathfac @Skylion007 @abhiskk for resolution issue.
from habitat-challenge.
commented on September 12, 2024
@abhiskk @mathfac @Skylion007 any suggestions ?
from habitat-challenge.
mathfac
commented on September 12, 2024
@AdventureO, we have PR in progress that will introduce crop and resize functionality to Habitat baselines: https://github.com/facebookresearch/habitat-api/pull/365/files. For now, you can change challenge config to square resolution to unblock your training.
from habitat-challenge.
commented on September 12, 2024
@mathfac now I get the following error:
File "ppo_agent.py", line 41, in <module>
main()
File "ppo_agent.py", line 37, in main
challenge.submit(agent)
File "/habitat-api/habitat/core/challenge.py", line 19, in submit
metrics = super().evaluate(agent)
File "/habitat-api/habitat/core/benchmark.py", line 163, in evaluate
return self.local_evaluate(agent, num_episodes)
File "/habitat-api/habitat/core/benchmark.py", line 137, in local_evaluate
action = agent.act(observations)
File "/habitat-api/habitat_baselines/agents/ppo_agents.py", line 134, in act
deterministic=False,
File "/habitat-api/habitat_baselines/rl/ppo/policy.py", line 40, in act
observations, rnn_hidden_states, prev_actions, masks
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
File "/habitat-api/habitat_baselines/rl/ppo/policy.py", line 167, in forward
perception_embed = self.visual_encoder(observations)
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
File "/habitat-api/habitat_baselines/rl/models/simple_cnn.py", line 147, in forward
return self.cnn(cnn_input)
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/container.py", line 100, in forward
input = module(input)
File "/opt/conda/envs/habitat/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
File "/habitat-api/habitat_baselines/common/utils.py", line 22, in forward
return x.view(x.size(0), -1)
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.
Does it make sense now to try to finally train and evaluate ppo model or better switch to dd-ppo baseline ?
from habitat-challenge.
mathfac
commented on September 12, 2024
@AdventureO, yes DDPPO baseline is preferred and merged into master facebookresearch/habitat-lab#370. Please, let me know if you face any issues running it.
from habitat-challenge.
commented on September 12, 2024
@mathfac to reduce training speed of DD-PPO I comented out all the lines which includes the key 'NOISE_MODEL' in config file as suggested, like this:
ENVIRONMENT:
MAX_EPISODE_STEPS: 500
ITERATOR_OPTIONS:
SHUFFLE: False
SIMULATOR:
TURN_ANGLE: 30
AGENT_0:
SENSORS: ['RGB_SENSOR', 'DEPTH_SENSOR']
HEIGHT: 0.88
RADIUS: 0.18
HABITAT_SIM_V0:
GPU_DEVICE_ID: 1
ALLOW_SLIDING: False
RGB_SENSOR:
WIDTH: 640
HEIGHT: 360
HFOV: 70
POSITION: [0, 0.88, 0]
# NOISE_MODEL: "GaussianNoiseModel"
# NOISE_MODEL_KWARGS:
# intensity_constant: 0.1
DEPTH_SENSOR:
WIDTH: 640
HEIGHT: 360
HFOV: 70
MIN_DEPTH: 0.1
MAX_DEPTH: 10.0
POSITION: [0, 0.88, 0]
# NOISE_MODEL: "RedwoodDepthNoiseModel"
# ACTION_SPACE_CONFIG: 'pyrobotnoisy'
# NOISE_MODEL:
# ROBOT: "LoCoBot"
# CONTROLLER: 'Proportional'
# NOISE_MULTIPLIER: 0.5
TASK:
TYPE: Nav-v0
SUCCESS_DISTANCE: 0.36 # 2 x Agent Radius
SENSORS: ['POINTGOAL_SENSOR']
POINTGOAL_SENSOR:
GOAL_FORMAT: POLAR
DIMENSIONALITY: 2
GOAL_SENSOR_UUID: pointgoal
MEASUREMENTS: ['DISTANCE_TO_GOAL', "SUCCESS", 'SPL']
SUCCESS:
SUCCESS_DISTANCE: 0.36 # 2 x Agent Radius
But I've got the following error:
File "habitat_baselines/run.py", line 74, in <module>
main()
File "habitat_baselines/run.py", line 41, in main
run_exp(**vars(args))
File "habitat_baselines/run.py", line 68, in run_exp
trainer.train()
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/rl/ddppo/algo/ddppo_trainer.py", line 355, in train
) = self._update_agent(ppo_cfg, rollouts)
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/rl/ppo/ppo_trainer.py", line 256, in _update_agent
value_loss, action_loss, dist_entropy = self.agent.update(rollouts)
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/rl/ppo/ppo.py", line 74, in update
for sample in data_generator:
File "/home/pryhoda/HabitatProject/habitat-api/habitat_baselines/common/rollout_storage.py", line 157, in recurrent_generator
ind = perm[start_ind + offset]
IndexError: index 7 is out of bounds for dimension 0 with size 7
Could you please help to fix it ? Probably there should some other value for ACTION_SPACE_CONFIG key.
from habitat-challenge.
erikwijmans
commented on September 12, 2024
You will certainly want to keep the noisy actions as that doesn't slow down simulation speed by much (if it all), i.e. these
# ACTION_SPACE_CONFIG: 'pyrobotnoisy'
# NOISE_MODEL:
# ROBOT: "LoCoBot"
# CONTROLLER: 'Proportional'
# NOISE_MULTIPLIER: 0.5
The second error is likely due to the number of processes vs. the number of PPO mini batches. The number of processes should really be an integer multiple of the number of PPO mini batches.
from habitat-challenge.
commented on September 12, 2024
@erikwijmans thanks!
Is there any ways to improve the speed of training for DD-PPO ?
Training procedure with 4 processes on GeForce RTX 2080ti is very slow
from habitat-challenge.
Skylion007
commented on September 12, 2024
Yes, there are ways to further speed up DD-PPO. You enable the
SIMULATOR.HABITAT_SIM_V0.GPU2GPU flag to allow for GPU2GPU transfers
(however, note that habitat-sim will need to be built --with-cuda and it
may reduce the amount of available VRAM on the card since each simulator
process will need to allocate memory to store the CUDA runtime.)
Also with the above error, the action space config should be set to v1 for
the agent if you want to the agent to be able use the look up and down
actions.
ACTION_SPACE_CONFIG: 'v1'
…On Tue, May 5, 2020 at 9:43 AM Oleksandr Pryhoda ***@***.***> wrote:
@erikwijmans <https://github.com/erikwijmans> thanks!
Is there any ways to improve the speed of training for DD-PPO ?
Training procedure with 4 processes on GeForce RTX 2080ti is very slow
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#26 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAPVMX5OWXBHKQUQ63Z6PVLRQAJZLANCNFSM4LTVHVNQ>
.
from habitat-challenge.
erikwijmans
commented on September 12, 2024
Looks like that code wasn't setup to read those parameters and needs to be modified to use resnet18 instead of resnet50
from habitat-challenge.
commented on September 12, 2024
@erikwijmans except config file, where else should I modify the code ?
from habitat-challenge.
mathfac
commented on September 12, 2024
@AdventureO #47 should fix it for local Docker evaluation.
from habitat-challenge.
Related Issues (20)
- distance_to_goal calculation results in infinity, pitch control is not working effectively in continuous action space HOT 5
- Custom Env class HOT 1
- Can not run pretrained weights of ObjectNav HOT 3
- Allow sliding HOT 1
- Settings for 2023
- submission killed by server
- Error when try to local evaluate : "6. Evaluate your docker container locally" HOT 2
- Could not submit a docker at EvalAI page HOT 4
- challenge 2023 HOT 1
- Inapproriate Settings about Continuous Action Space in habitat-chllenge 2023
- Eval.ai submission is not evaluating HOT 2
- deadline of challenge 2023
- miscalculated the deadline
- wrong deadline setting on hosts of challenge 2023!!!
- wrong deadline setting on hosts of challenge 2023!!!
- wrong deadline setting on hosts of challenge 2023!!! HOT 10
- The final list of the challenge phase HOT 3
- ConnectionResetError: [Errno 104] Connection reset by peer HOT 1
- AssertionError: Object attributes not uniquely matched to shortened handle. HOT 1
- AttributeError: Can't get attribute 'VectorEnvFactoryConfig'
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from habitat-challenge.