kelym / fast Goto Github PK
View Code? Open in Web Editor NEWCode for "Tactical Rewind: Self-Correction via Backtracking in Vision-and-Language Navigation"
Home Page: https://arxiv.org/abs/1903.02547
Code for "Tactical Rewind: Self-Correction via Backtracking in Vision-and-Language Navigation"
Home Page: https://arxiv.org/abs/1903.02547
Hello, thank u for sharing your code.
I want to duplicate your model performances reported in your paper.
(Table 1- 9nd row: SMNA + FAST (beam) , 6nd row : SMNA + FAST (Greedy))
Would you please tell me how to duplicate the performance?
And I followed your readme(6th command).
It made two files(cache_val_unseen40False.json, search_val_unseen40False.json).
And I ran "Training Reranker.ipynb" file ,but i encountered some problem.
"FileNotFoundError: [Errno 2] No such file or directory: 'cache_train40False.json'"
There is a line about this in .ipynb
[train_data, val_seen, val_unseen] = load_data(['cache_train40False.json','cache_val_seen40False.json','cache_val_unseen40False.json'])
I think this code needs additional files(train40False.json file, val_seen40False.json)
How to get these file?
Where i should modify? ==> I fix this problem by modify the run_search.py.
Now, I got cache_train40False.json, cache_val_seen40False.json, cache_val_unseen40False.json, search_train40False.json, search_val_seen40False.json, search_val_unseen40False.json
But, the another problem still occurs.
I got issue on (line #40) Training Reranker.ipynb(below)
_, world_states, actions, sum_logits, mean_logits, sum_logp, mean_logp, pm, speaker, scorer = candidate
ValueError: not enough values to unpack (expected 10, got 4)
I printed the candidate(value) but, It only includes "traj"
How do i fix this problem?
Please answer to me.
Thank you!
Hi. I cant download the http://people.eecs.berkeley.edu/~ronghang/projects/speaker_follower/data_augmentation/R2R_literal_speaker_data_augmentation_paths.json
from download_precomputed_augmentation.sh
. Can you repair the link?
Thank you
Hi,
Can you explain where the backtracking code is? I read that in follwer.py, the _rollout_with_search function uses a greedy decoding method. Is that correct?
Thanks.
When I execute wget https://storage.googleapis.com/bringmeaspoon/R2Rdata/R2R_train.json -P tasks/R2R/data/, an error will be returned as Anonymous caller does not have storage.objects.get access to bringmeaspoon/R2Rdata/R2R_train.json
Hi, this is Hualin. Thanks for sharing your code. I am trying to reproduce your results using the steps in README.md. I encountered several errors and hope you could help me with it.
I notice that the Matterport3D repo has changed their c++ api. Hence, I couldn't directly copy your code to tasks/R2R and run it. Could you give me the exact version number of their repo based on which you conduct your experiment?
After I found the above mentioned repo API problem I switched to the the Regretful Agent repo (Another work from the same author of SMNA). It seemed they used the old api and I got the training procedure running. Then another bug arose. After I trained the agent for one iteration using first command in step 5, some errors happened in the evaluation step. The error logs are as follows:
Training with sample2step feedback
100%|███████████████████████████████████████████████████████████████████████████████| 100/100 [02:33<00:00, 1.19s/it]
/opt/conda/lib/python3.6/site-packages/numpy/lib/function_base.py:392: RuntimeWarning: Mean of empty slice.
avg = a.mean(axis)
/opt/conda/lib/python3.6/site-packages/numpy/core/_methods.py:85: RuntimeWarning: invalid value encountered in double_scalars
ret = ret.dtype.type(ret / rcount)
evaluating on val_seen
Traceback (most recent call last):
File "tasks/R2R/train.py", line 378, in
utils.run(make_arg_parser(), train_val)
File "/workspace/selfmonitoring-agent/tasks/R2R/utils.py", line 315, in run
entry_function(args)
File "tasks/R2R/train.py", line 321, in train_val
args.n_pretrain_iters, val_envs=val_envs)
File "tasks/R2R/train.py", line 125, in train
score_summary, _ = evaluator.score_results(agent.results)
File "/workspace/selfmonitoring-agent/tasks/R2R/eval.py", line 144, in score_results
score_summary['model_score'] = np.average(model_scores)
File "/opt/conda/lib/python3.6/site-packages/numpy/lib/function_base.py", line 392, in average
avg = a.mean(axis)
File "/opt/conda/lib/python3.6/site-packages/numpy/core/_methods.py", line 85, in _mean
ret = ret.dtype.type(ret / rcount)
AttributeError: 'torch.dtype' object has no attribute 'type'
May I know if you have any insights on what could possibly cause this? Thank you and looking forward to your reply.
Thank you for sharing nice work.
I have trouble in training.
While my model trained with below command, my computer suddenly shut down.
python tasks/R2R/train.py --use_pretraining
--pretrain_splits train literal_speaker_data_augmentation_paths
--feedback_method sample2step
--experiment_name smna
I checked the process and saved model file, it was already finished training with augmented data and was getting process with real data.
I want to resuming training with real data at that final shut down timestep(epoch).
What could i do to resume the training?
hello.
I'm sorry to raise another issue.
I have problem with final command.
python tasks/R2R/run_search.py --job search --load_follower tasks/R2R/experiments/smna/snapshots/[name of the latest model] --max_episode_len 55 --K 20 --logit --experiment_name tmp --load_reranker tasks/R2R/experiments/[name of the reranker model]
I ran Training reranker.ipynb and got new candidates_ranker_xxxx models.
but, after that i ran your command, got error like below.
I think these caused by this line in your .ipynb file.
I don't know why you did the input_dim size to 7. Could you please explain about this?
And how do i fix the problem?
I'll be waiting for the reply.
Thank you.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.