Fix and refactor SAC

docs/homepage/blog/a_practical_introduction_to_RL.jl/index.html

@@ -15129,7 +15129,7 @@ <h2 id="Two-Most-Commonly-Used-Algorithms">Two Most Commonly Used Algorithms<a c
                 <span class="p">)</span> <span class="o">|&gt;</span> <span class="n">cpu</span><span class="p">,</span>
                 <span class="n">optimizer</span> <span class="o">=</span> <span class="n">Adam</span><span class="p">(),</span>
             <span class="p">),</span>
-            <span class="n">batch_size</span> <span class="o">=</span> <span class="mi">32</span><span class="p">,</span>
+            <span class="n">batchsize</span> <span class="o">=</span> <span class="mi">32</span><span class="p">,</span>
             <span class="n">min_replay_history</span> <span class="o">=</span> <span class="mi">100</span><span class="p">,</span>
             <span class="n">loss_func</span> <span class="o">=</span> <span class="n">huber_loss</span><span class="p">,</span>
             <span class="n">rng</span> <span class="o">=</span> <span class="n">rng</span><span class="p">,</span>

docs/homepage/blog/ospp_final_term_report_210370741/index.md

@@ -314,7 +314,7 @@ struct D4RLDataSet{T<:AbstractRNG} <: RLDataSet
     dataset::Dict{Symbol, Any}
     repo::String
     dataset_size::Integer
-    batch_size::Integer
+    batchsize::Integer
     style::Tuple
     rng::T
     meta::Dict
@@ -330,7 +330,7 @@ function dataset(dataset::String;
     repo = "d4rl",
     rng = StableRNG(123), 
     is_shuffle = true, 
-    batch_size=256
+    batchsize=256
 )
 ```
 
@@ -383,7 +383,7 @@ Multi threaded batching using a parallel loop where each thread loads the batche
 
 ```julia
 res = Channel{AtariRLTransition}(n_preallocations; taskref=taskref, spawn=true) do ch
-    Threads.@threads for i in 1:batch_size
+    Threads.@threads for i in 1:batchsize
         put!(ch, deepcopy(batch(buffer_template, popfirst!(transitions), i)))
     end
 end
@@ -472,7 +472,7 @@ end
 The datapoints are then put in a `RingBuffer` which is returned.
 ```julia
 res = RingBuffer(buffer;taskref=taskref, sz=n_preallocations) do buff
-    Threads.@threads for i in 1:batch_size
+    Threads.@threads for i in 1:batchsize
         batch!(buff, take!(transitions), i)
     end
 end
@@ -694,7 +694,7 @@ mutable struct FQE{
     target_q_network::C_T
     n_evals::Int
     γ::Float32
-    batch_size::Int
+    batchsize::Int
     update_freq::Int
     update_step::Int
     tar_update_freq::Int
@@ -714,7 +714,7 @@ function RLBase.update!(l::FQE, batch::NamedTuple{SARTS})
     D = device(Q)
     s, a, r, t, s′ = (send_to_device(D, batch[x]) for x in SARTS)
     γ = l.γ
-    batch_size = l.batch_size
+    batchsize = l.batchsize
 
     loss_func = Flux.Losses.mse
 
@@ -723,7 +723,7 @@ function RLBase.update!(l::FQE, batch::NamedTuple{SARTS})
     target = r .+ γ .* (1 .- t) .* q′
 
     gs = gradient(params(Q)) do
-        q = Q(vcat(s, reshape(a, :, batch_size))) |> vec
+        q = Q(vcat(s, reshape(a, :, batchsize))) |> vec
         loss = loss_func(q, target)
         Zygote.ignore() do
             l.loss = loss

docs/homepage/blog/ospp_report_210370190/index.md

@@ -247,7 +247,7 @@ rl_agent = Agent(
             ),
             γ = 1.0f0,
             loss_func = huber_loss,
-            batch_size = 128,
+            batchsize = 128,
             update_freq = 128,
             min_replay_history = 1000,
             target_update_freq = 1000,
@@ -281,7 +281,7 @@ sl_agent = Agent(
             optimizer = Descent(0.01),
         ),
         explorer = WeightedSoftmaxExplorer(),
-        batch_size = 128,
+        batchsize = 128,
         min_reservoir_history = 1000,
         rng = rng,
     ),
@@ -351,7 +351,7 @@ Given that the [`DDPGPolicy`](https://juliareinforcementlearning.org/docs/rlzoo/
 mutable struct MADDPGManager <: AbstractPolicy
     agents::Dict{<:Any, <:Agent}
     traces
-    batch_size::Int
+    batchsize::Int
     update_freq::Int
     update_step::Int
     rng::AbstractRNG
@@ -454,7 +454,7 @@ agents = MADDPGManager(
         trajectory = deepcopy(trajectory),
     )) for player in players(env) if player != chance_player(env)),
     SARTS, # trace's type
-    512, # batch_size
+    512, # batchsize
     100, # update_freq
     0, # initial update_step
     rng
@@ -508,7 +508,7 @@ create_policy(player) = DDPGPolicy(
     na = length(action_space(env, player)),
     start_steps = 0,
     start_policy = nothing,
-    update_after = 512 * env.max_steps, # batch_size * env.max_steps
+    update_after = 512 * env.max_steps, # batchsize * env.max_steps
     act_limit = 1.0,
     act_noise = 0.,
     )
@@ -530,7 +530,7 @@ agents = MADDPGManager(
         ) for player in (:Speaker, :Listener)
     ),
     SARTS, # trace's type
-    512, # batch_size
+    512, # batchsize
     100, # update_freq
     0, # initial update_step
     rng

docs/homepage/blog/phase1_technical_report_of_enriching_offline_reinforcement_learning_algorithms_in_reinforcement_learning_jl/index.md

@@ -57,10 +57,10 @@ Base.@kwdef struct OfflinePolicy{L,T} <: AbstractPolicy
     learner::L
     dataset::T
     continuous::Bool
-    batch_size::Int
+    batchsize::Int
 end
 ```
-This implementation of `OfflinePolicy` refers to [`QBasePolicy`](https://juliareinforcementlearning.org/docs/rlcore/#ReinforcementLearningCore.QBasedPolicy). It provides a parameter `continuous` to support different action space types, including continuous and discrete. `learner` is a specific algorithm for learning and providing policy. `dataset` and `batch_size` are used to sample data for learning.
+This implementation of `OfflinePolicy` refers to [`QBasePolicy`](https://juliareinforcementlearning.org/docs/rlcore/#ReinforcementLearningCore.QBasedPolicy). It provides a parameter `continuous` to support different action space types, including continuous and discrete. `learner` is a specific algorithm for learning and providing policy. `dataset` and `batchsize` are used to sample data for learning.
 
 Besides, we implement corresponding functions `π`, `update!` and `sample`. `π` is used to select the action, whose form is determined by the type of action space. `update!` can be used in two stages. In `PreExperiment` stage, we can call this function for pre-training algorithms with `pretrain_step` parameters. In `PreAct` stage, we call this function for training the `learner`. In function `update!`, we need to call function `sample` to sample a batch of data from the dataset. With the development of [ReinforcementLearningDataset.jl](https://github.com/JuliaReinforcementLearning/ReinforcementLearning.jl/tree/master/src/ReinforcementLearningDatasets), the `sample` function will be deprecated.
 
@@ -73,7 +73,7 @@ offline_dqn_policy = OfflinePolicy(
     ),
     dataset = dataset,
     continuous = false,
-    batch_size = 64,
+    batchsize = 64,
 )
 ```
 
@@ -266,7 +266,7 @@ function RLBase.update!(p::OfflinePolicy, traj::AbstractTrajectory, ::AbstractEn
     if in(:pretrain_step, fieldnames(typeof(l)))
         println("Pretrain...")
         for _ in 1:l.pretrain_step
-            inds, batch = sample(l.rng, p.dataset, p.batch_size)
+            inds, batch = sample(l.rng, p.dataset, p.batchsize)
             update!(l, batch)
         end
     end

docs/src/How_to_use_hooks.md

@@ -138,7 +138,7 @@ policy = Agent(
                 ) |> cpu,
                 optimizer = Adam(),
             ),
-            batch_size = 32,
+            batchsize = 32,
             min_replay_history = 100,
             loss_func = huber_loss,
         ),

src/ReinforcementLearningBase/Project.toml

@@ -1,7 +1,7 @@
 name = "ReinforcementLearningBase"
 uuid = "e575027e-6cd6-5018-9292-cdc6200d2b44"
 authors = ["Johanni Brea <[email protected]>", "Jun Tian <[email protected]>"]
-version = "0.12.1"
+version = "0.12.2"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"

src/ReinforcementLearningCore/Project.toml

@@ -1,6 +1,6 @@
 name = "ReinforcementLearningCore"
 uuid = "de1b191a-4ae0-4afa-a27b-92d07f46b2d6"
-version = "0.14.0"
+version = "0.15"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"
@@ -30,7 +30,7 @@ cuDNN = "02a925ec-e4fe-4b08-9a7e-0d78e3d38ccd"
 [compat]
 AbstractTrees = "0.3, 0.4"
 Adapt = "3"
-CUDA = "4"
+CUDA = "4, 5"
 ChainRulesCore = "1"
 CircularArrayBuffers = "0.1"
 Crayons = "4"

src/ReinforcementLearningCore/src/core/hooks.jl

@@ -217,13 +217,13 @@ end
 Base.getindex(h::BatchStepsPerEpisode) = h.steps
 
 """
-    BatchStepsPerEpisode(batch_size::Int; tag = "TRAINING")
+    BatchStepsPerEpisode(batchsize::Int; tag = "TRAINING")
 
 Similar to [`StepsPerEpisode`](@ref), but is specific to environments
 which return a `Vector` of rewards (a typical case with `MultiThreadEnv`).
 """
-function BatchStepsPerEpisode(batch_size::Int)
-    BatchStepsPerEpisode([Int[] for _ = 1:batch_size], zeros(Int, batch_size))
+function BatchStepsPerEpisode(batchsize::Int)
+    BatchStepsPerEpisode([Int[] for _ = 1:batchsize], zeros(Int, batchsize))
 end
 
 function Base.push!(hook::BatchStepsPerEpisode, 

src/ReinforcementLearningCore/src/utils/distributions.jl

@@ -54,10 +54,10 @@ end
     mvnormlogpdf(μ::A, LorU::A, x::A; ϵ = 1f-8) where A <: AbstractArray
 
 Batch version that takes 3D tensors as input where each slice along the 3rd
-dimension is a batch sample.  `μ` is a (action_size x 1 x batch_size) matrix,
-`L` is a (action_size x action_size x batch_size), x is a (action_size x
-action_samples x batch_size).  Return a 3D matrix of size (1 x action_samples x
-batch_size). 
+dimension is a batch sample.  `μ` is a (action_size x 1 x batchsize) matrix,
+`L` is a (action_size x action_size x batchsize), x is a (action_size x
+action_samples x batchsize).  Return a 3D matrix of size (1 x action_samples x
+batchsize). 
 """
 function mvnormlogpdf(μ::A, LorU::A, x::A; ϵ=1.0f-8) where {A<:AbstractArray}
     it = zip(eachslice(μ, dims = 3), eachslice(LorU, dims = 3), eachslice(x, dims = 3))