refactor: update configuration_tdmpc.py (#153)

Co-authored-by: Alexander Soare <alexander.soare159@gmail.com>
这个提交包含在:
Ikko Eltociear Ashimine
2024-05-09 02:13:51 +09:00
提交者 GitHub
父节点 460df2ccea
当前提交 0ea7a8b2a3

查看文件

@ -47,7 +47,7 @@ class TDMPCConfig:
elite_weighting_temperature: The temperature to use for softmax weighting (by trajectory value) of the elite_weighting_temperature: The temperature to use for softmax weighting (by trajectory value) of the
elites, when updating the gaussian parameters for CEM. elites, when updating the gaussian parameters for CEM.
gaussian_mean_momentum: Momentum (α) used for EMA updates of the mean parameter μ of the gaussian gaussian_mean_momentum: Momentum (α) used for EMA updates of the mean parameter μ of the gaussian
paramters optimized in CEM. Updates are calculated as μ⁻ ← αμ⁻ + (1-α)μ. parameters optimized in CEM. Updates are calculated as μ⁻ ← αμ⁻ + (1-α)μ.
max_random_shift_ratio: Maximum random shift (as a proportion of the image size) to apply to the max_random_shift_ratio: Maximum random shift (as a proportion of the image size) to apply to the
image(s) (in units of pixels) for training-time augmentation. If set to 0, no such augmentation image(s) (in units of pixels) for training-time augmentation. If set to 0, no such augmentation
is applied. Note that the input images are assumed to be square for this augmentation. is applied. Note that the input images are assumed to be square for this augmentation.