-
Notifications
You must be signed in to change notification settings - Fork 1
/
1448160--input.jl
145 lines (123 loc) · 4.3 KB
/
1448160--input.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
algo="DDPG"
# Parameters and architecture based on:
# https://github.com/msinto93/DDPG/blob/master/train.py
# https://github.com/JuliaReinforcementLearning/ReinforcementLearningZoo.jl/blob/master/src/experiments/rl_envs/JuliaRL_DDPG_Pendulum.jl
# https://github.com/FluxML/model-zoo/blob/master/contrib/games/differentiable-programming/pendulum/DDPG.jl
using Pkg
Pkg.activate(pwd())
using Flux, Printf, Zygote
using Flux.Optimise: update!
using BSON
using Statistics: mean, std, median
using DataStructures: CircularBuffer
using Distributions: sample, Normal, logpdf
using Random
using Reinforce
using Dates
using Plots
using CSV, DataFrames
gr()
#------------ local machine ----------
# Job_ID=1149869
# Task_ID = 1149869-1 #ENV["SGE_TASK_ID"]
# seed_run = 10 # run inference over all seeds
#--------cluster jobs------------
Job_ID = ENV["JOB_ID"]
Task_ID = ENV["SGE_TASK_ID"]
seed_run = parse(Int, Task_ID)
#-------------------------------- INPUTS --------------------------------------------
train = 0
plot_result = 0
plot_all = 0
render = 0
track = -0.6 # 0 - off, 1 - DRL, -1 - rule-based 1, -2 rule-based 2
season = "all" # "all" "both" "summer" "winter"
price= "fix" # "fix", "TOU"
noise_type = "gn" # "ou", "pn", "gn", "en"
using Reinforce.ShemsEnv_U8: Shems
case = "$(season)_$(algo)_$(price)_base-256_gn.1_Env-U8-no-layer-norm"
run = "eval"
NUM_EP = 3_001 #50_000
L1 = 300 #256
L2 = 600 #256
idx=NUM_EP
test_every = 100
test_runs = 100
num_seeds = 40
#-------------------------------------
seed_ini = 123
# individual random seed for each run
rng_run = parse(Int, string(seed_ini)*string(seed_run))
start_time = now()
current_episode = 0
#--------------------------------- Memory ------------------------------------
BATCH_SIZE = 120 #100 # Yu: 120
MEM_SIZE = 24_000
MIN_EXP_SIZE = 24_000
########################################################################################
memory = CircularBuffer{Any}(MEM_SIZE)
#--------------------------------- Game environment ---------------------------
EP_LENGTH = Dict("train" => 24,
("summer", "eval") => 359, ("summer", "test") => 767,
("winter", "eval") => 359, ("winter", "test") => 719,
("both", "eval") => 719, ("both", "test") => 1487,
("all", "eval") => 1439, ("all", "test") => 2999) # length of whole evaluation set (different)
env_dict = Dict("train" => Shems(EP_LENGTH["train"], "data/$(season)_train_$(price).csv"),
"eval" => Shems(EP_LENGTH[season, "eval"], "data/$(season)_eval_$(price).csv"),
"test" => Shems(EP_LENGTH[season, "test"], "data/$(season)_test_$(price).csv"))
WAIT = Dict(
("summer", "DDPG") => 1500, ("summer", "TD3") => 1500, ("summer", "SAC") => 5000,
("winter", "DDPG") => 2000, ("winter", "TD3") => 2000, ("winter", "SAC") => 7000,
("both", "DDPG") => 3000, ("both", "TD3") => 3000, ("both", "SAC") => 10000,
("all", "DDPG") => 4000, ("all", "TD3") => 5000, ("all", "SAC") => 20_000)
# ----------------------------- Environment Parameters -------------------------
STATE_SIZE = length(env_dict["train"].state)
ACTION_SIZE = length(env_dict["train"].a)
ACTION_BOUND_HI = maximum(env_dict["train"].a)
ACTION_BOUND_LO = minimum(env_dict["train"].a)
# ACTION_BOUND_HI = Float32[1f0, 1f0, 1f0] #Float32(actions(env, env.state).hi[1])
# ACTION_BOUND_LO = Float32[-1f0, -1f0, -1f0] #Float32(actions(env, env.state).lo[1])
# ACTION_BOUND_HI = Float32[4.6f0, 3.0f0] #Float32(actions(env, env.state).hi[1])
# ACTION_BOUND_LO = Float32[-4.6f0, -3.0f0] #Float32(actions(env, env.state).lo[1])
#------------------------------- Action Noise --------------------------------
struct OUNoise
μ
σ
θ
dt
X
end
struct GNoise
μ
σ_act
σ_trg
end
mutable struct EpsNoise
ζ
ξ
ξ_min
end
mutable struct ParamNoise
μ
σ_current
σ_target
adoption
end
# Ornstein-Uhlenbeck / Gaussian Noise params
# based on: https://github.com/openai/baselines/blob/master/baselines/ddpg/noise.py
μ = 0f0 #mu
σ = 0.1f0 #0.2f0 #sigma
θ = 0.15f0 #theta
dt = 1f-2
# Epsilon Noise parameters based on Yu et al. 2019
ζ = 0.0005f0
ξ_0 = 0.5f0
ξ_min = 0.1f0
# Noise actor
noise_act = 0.1f0 #0.1f0 #2f-1
noise_trg = 0.2f0 #3f-1
# Fill struct with values
ou = OUNoise(μ, σ, θ, dt, zeros(Float32, ACTION_SIZE))
gn = GNoise(μ, noise_act, noise_trg)
en = EpsNoise(ζ, ξ_0, ξ_min)
pn = ParamNoise(μ, σ, noise_act, 1.01)