DI-star VS pytorch-lightning

There is an AI/bot scene for SC2, I don't have many links but you can start by looking here: https://github.com/opendilab/DI-star https://www.youtube.com/watch?v=fvQF-24IpXs (Harstem and uThermal both have more videos vs different bots).

Looks like a simplified AlphaStar using LSTM RNN instead of Pointer Transformer, much heavier supervised imitation learning, Zerg vs Zerg only (with simplified build order module), and a much smaller AlphaStar League: https://github.com/opendilab/DI-star/blob/main/docs/guidance_to_small_scale_training.md

For more information，plz visit out GitHub page：https://github.com/opendilab/DI-star

First of all, let's discuss what the level of AI is now. If the "level" refers to the capability of competing, the current AI has been very closed to the top human player in some types of games, like chess, Texas Poker, and Mahjong of CARDS, DOTA2 of MOBA, as well as StarCraft2 of RTS. As for other games, if we have enough human resources and computing performance, we also can get similar results. If the "level" has other meanings, like AI agents having human behavior, intelligent NPC can be designed specifically for different people so that they can have different gaming experience. These are all at the stage of issue-defining and exploring new technology solutions. Although traditional game AI is mostly based on hard code, it still has much prior knowledge. In recent years, some hot ML-related techs have performed well in competitiveness while in other fields, they haven't found the perfect entry point. If we expand the conclusions above in detail, the design of game AI can be divided into two parts: issue defining and issue solving. For those competitive issues which have already got complete definitions, their core issue is to explore the optimal strategy based on the evaluation standard, like ladder points. Traditional solutions can deal with less complex scenarios, like chess and Gobang. While machine learning related techs, including deep learning and reinforcement learning, they can perform very well in much more complex games, like StarCraft II. {For this you can try it in DI-star: this project is a reimplementation (with a few improvements) of Alphastar (Only Zerg vs Zerg) based on OpenDILab.}

def on_validation_end(self, trainer: Trainer, pl_module: LightningModule) -> None: # Trainer calls `on_validation_end` for sanity check. Therefore, it is necessary to avoid # calling `trial.report` multiple times at epoch 0. For more details, see # https://github.com/PyTorchLightning/pytorch-lightning/issues/1391. if trainer.sanity_checking: return

Under the hood, the ModelWrapper object will create a ML model based on the config (so far, an XGBoost model and a PyTorch Lightning model). Each of those will have a wrapper that conducts training and evaluation (since from my understanding of Lightning, Trainers are required to be outside of the class). In lack of a better name, I call these wrappers Fitters. For uniformity, I thought about adding a common interface IFitter, which is inherited by all model wrappers as outlined below.

Join their Slack to ask the community questions and check out the GitHub here.

Pytorch lightning benchmarks against pytorch on every PR (benchmarks to make sure that it is mot slower.

There is already a ton of momentum around automating ML workflows. I would suggest you contribute to a preexisting project like, for instance, PyTorch Lightning or fast.ai.

Is the problem the sheer number of options, or the fact that they are all together in one place? Would it be better if they were organized into the different trainer entrypoints (fit, validate, ...)? If that is the case, there was an RFC proposing this which you might find interesting, feel free to drop by and comment on the issue: https://github.com/PyTorchLightning/pytorch-lightning/issues/10444

I wanted to make a quick performance comparison between the GPU (Tesla K80) and TPU (v2-8) available in Google Colab with PyTorch. To do so quickly, I used an MNIST example from pytorch-lightning that trains a simple CNN.

We've noticed GPU 0 on our 3 GPU system is sometimes idle (which would explain performance differences). However its unclear to us why that may be. Similar to this issue

If you want a good example for reference, take a look at Pytorch Lightning's readme (https://github.com/PyTorchLightning/pytorch-lightning) It answers the 3 questions of "what is this", "why should I care", and "how do i use it" almost instantly