老董

On win10, torch 1.11 and tch-rs 0.7.2: ```rust let tensor_to_transform = Tensor::try_from(array![ [0.2f32, 1.0, 0.0], [0.0, 0.9, -0.14], [0.0, -0.2, 1.0] ]) .unwrap(); let mut tensor_to_transform_2 = Tensor::try_from(array![ [0.2f32, 1.0,...

This API could be found in 0.7, but not 0.8. Why delete it? Is there some replaced API for it?

I use tch-rs "0.7.2" with `libtorch-win-shared-with-deps-1.11.0+cu113` on win10. Sometimes, within the output of this layer: ```rust let shared_feature_layer = nn::seq() .add(nn::linear( p / "actor_feature", input_len, 128, Default::default(), )) .add_fn(|xs| xs.relu());...

Release mode cannot find cuda

2

On Win10 with cuda 11.3, and tch-rs 0.7.2 and [torchlib 1.11.0_cuda11.3](https://pytorch.org/get-started/locally/) of release version, when calling ` tch::Device::cuda_if_available(),` in release mode, it turns to using cpu. If it is forced...

There are parameter `alpha_log_d` and related optimizer `alpha_d_opt`: ```rust let alpha_log_d = alpha_vs_d.root().var( "log_alpha_d", &[1], nn::Init::Const(-action_dim_len.ln() * f64::exp(1f64)), ); let mut alpha_d_opt = nn::Adam::default() .build(&alpha_vs_d, opt_learning_rate) .expect("init nn::Adam failed for...

What is the best way to save tensor to disk?

2

When doing with reinforcement learning project, I have a struct like this: ```rust Struct { obs: vec, action: vec, ... } ``` I need to save the whole instance of...

Is it possible to use ModuleDict?

2

For a more flexible torch network, sometimes [ModuleDicts](https://pytorch.org/docs/stable/generated/torch.nn.ModuleDict.html) would be used. I saw the c++ api is release since [pytorch 1.8](https://github.com/pytorch/pytorch/releases?after=v1.8.1-rc1), [related issue](https://github.com/pytorch/pytorch/pull/47707), but I could not find related document...

Since COMBO is derived from CQL, I just wonder why the auto adjustment used in CQL is not discussed in COMBO? Is it useless in COMBO?

1.For behavior cloning, the update formula ` policy_loss = (alpha*log_pi - log_probs).mean()`, I wonder why using `log_probs` , but not q-value here? 2. When using Lagrange, do `alpha_prime` and `cql_min_q_weight`...

有示范吗？