Metadata-Version: 2.1
Name: torchlure
Version: 0.2405.7
Author-email: fuyutarow <fuyutarow@gmail.com>
Project-URL: Homepage, https://github.com/fuyutarow/torchlure
Project-URL: Repository, https://github.com/fuyutarow/torchlure
Keywords: pytorch
Requires-Python: >=3.8
Description-Content-Type: text/markdown
Requires-Dist: torch>=2.1
Requires-Dist: numpy>=1.26.4
Requires-Dist: sophia-opt>=0.2.2
Requires-Dist: gymnasium>=0.29.1
Requires-Dist: jax>=0.4.28
Requires-Dist: gdown>=5.2.0

# Torch Lure


<a href="https://www.youtube.com/watch?v=wCzCOYCfY9g" target="_blank">
  <img src="http://img.youtube.com/vi/wCzCOYCfY9g/maxresdefault.jpg" alt="Chandelure" style="width: 100%;">
</a>


# Depndencies

```
pip install git+https://github.com/Farama-Foundation/Minari.git@19565bd8cd33f2e4a3a9a8e4db372044b01ea8d3
```


```sh
pip install torchlure
```

# Usage
```py
import torchlure as lure

# Optimizers
lure.SophiaG(lr=1e-3, weight_decay=0.2)

# Functions
lure.tanh_exp(x)
lure.TanhExp()

lure.quantile_loss(y_pred, y_target, quantile=0.5)
lure.QuantileLoss(quantile=0.5)

lure.RMSNrom(dim=256, eps=1e-6)

# Noise Scheduler
lure.LinearNoiseScheduler(beta=1e-4, beta_end=0.02, num_timesteps=1000)
lure.CosineNoiseScheduler(max_beta=0.999, s=0.008, num_timesteps=1000):
```

### Dataset



```py
from torchlure.datasets import OfflineRLDataset, D4RLDataset

# Initial usage with download
dataset = D4RLDataset(
    dataset_id="d4rl_walker2d-expert-2405",
    d4rl_name="d4rl_walker2d-expert-v2",
    env_id= "Walker2d-v4",
)

# if you are download it once
dataset = D4RLDataset(
    dataset_id="d4rl_walker2d-expert-2405",
)
```

See all datasets [here](https://github.com/pytorch/rl/blob/3a7cf6af2a08089f11e0ed8cad3dd1cea0e253fb/torchrl/data/datasets/d4rl_infos.py)
