Loading
0 Follower
0 Following
kaixin

Location

SG

Badges

1
0
0

Activity

Jun
Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Mon
Wed
Fri

Ratings Progression

Loading...

Challenge Categories

Loading...

Challenges Entered

Measure sample efficiency and generalization in reinforcement learning using procedurally generated environments

Latest submissions

See All
graded 76834
graded 76768
graded 75308

Sample-efficient reinforcement learning in Minecraft

Latest submissions

No submissions made in this challenge.

A new benchmark for Artificial Intelligence (AI) research in Reinforcement Learning

Latest submissions

No submissions made in this challenge.
Participant Rating
Participant Rating

NeurIPS 2020: Procgen Competition

Different training steps shown in gitlab

Almost 4 years ago

Out of the four games, it is shown that some games (e.g. coinrun, bigfish) train for 2.5M timesteps while only miner trains for 8M timesteps (see submission #75308). Is it a plotting bug or the numbers of training steps are actually different?

Customized trainer

Almost 4 years ago

@jyotish by the way, can you help look at submission #75308? I just clone and submit the impala-baseline but training failed in one of the environments.

Customized trainer

Almost 4 years ago

Thank you for your quick reply. I will look at them :smile:

Customized trainer

Almost 4 years ago

Can you give an example of customizing trainer? It is difficult to make large modifications via yaml file. Currently it seems that the start-kit only supports customizing envs and NN models.

Thank you.

Are entries on the leaderboard in round 1or warmup round?

Almost 4 years ago

Thank you for the quick reply!

Are entries on the leaderboard in round 1or warmup round?

Almost 4 years ago

I submitted a baseline model. It seems it is only evaluated on coinrun env.

NeurIPS 2019 : MineRL Competition

When will the results of round 1 be announced?

Over 4 years ago

The challenge page says 4th Nov. but it is already 5 Nov.

Evaluation waiting takes too long

Over 4 years ago

I note that it happens when training using multiple actors (one of instances hangs) and the error message is HTTPSConnectionPool(host='gitlab.aicrowd.com', port=443): Read timed out. (read timeout=None). I met this problem before, but after a re-queuing, the problem no longer exists and the evaluation goes smoothly.

Problems running in docker

Over 4 years ago

Same error here. I found this error happened if env.close() was called.

Evaluation waiting takes too long

Over 4 years ago

It has been 8 hours since the submission was tagged “evaluation initiated” but still no updates on testing running info. Is it normal?

Train a metacontroller based on Expert Data which is divided into several stages manually by reward

Over 4 years ago

I guess it is not allowed. The reward is closely related to the inventory. So using reward to split stages is equivalent to using inventory observation.

Using open-sourced networks

Over 4 years ago

@BrandonHoughton May I ask if you allow the use of network pre-trained on ImageNet considering the wide use of ImageNet pretraining? Thank you.

Best,
Kaixin

Repeated reward for logs and furnace

Over 4 years ago

I rechecked just now and also faced the auto-equipping issue. But weird things happened. This time I could not equip the wooden pickaxe. I tried several times but still could not get the wooden pickaxe equipped.

How to use furnace to "nearbySmelt" coal?

Over 4 years ago

Thanks for your reply but I am afraid you didn’t get my point. The minerl doc mentions an action nearbySmelt[coal] which I think means using the furnace (and some ingredients) to produce coal. And in fact, in the MineCraft game, it is possible (we can burn logs into charcoal).

Repeated reward for logs and furnace

Over 4 years ago

Yes, I built it myself. In ObtainDiamond env.

Yes, I can equip the pickaxe. I checked out the other thread about the equipping bug. I also met the same bug but it is slightly different. Once I have equipped the wood_pickaxe, I could not un-equip it by performing “equip air” action.

Repeated reward for logs and furnace

Over 4 years ago

I find a bug in minerl 0.2.4. The agent is rewarded every time when the quantity of furnace or log changes from 0 to 1. For example, if I use up logs to craft planks and then chop another tree, there will be another 1.0 reward. Similarly, if I place the furnace and collect it again by breaking it, I can repeatedly get 32.0 reward.

How to use furnace to "nearbySmelt" coal?

Over 4 years ago

I am new to MineCraft. May I ask what the ingredients of this action are? I tried with logs and planks in inventory but nothing happens. According to the game wiki, this should produce charcoal (using planks as fuel). Thanks.

Equip item failed

Over 4 years ago

I also met the same bug after upgrading to 0.2.4. I think it is a bug in the new version. Hope @BrandonHoughton can help.

Evaluation environment

Almost 5 years ago

@BrandonHoughton It is a bug indeed. For example, using the given test.py for evaluation, the agent has no “craft” action but it can obtain rewards up to 8 or 10 for some episode, which is impossible.

Evaluation environment

Almost 5 years ago

It seems the server is using ObtainDiamondDense env for evaluation instead of the sparse counterpart. Is that a bug?

Thank you

kaixin has not provided any information yet.