msaroufim

# https://github.com/JuliaML/Reinforce.jl/blob/master/examples/mountain_car.jl
using Reinforce
using Reinforce.MountainCarEnv: MountainCar

using Plots
gr()

# Deterministic policy that is solving the problem
mutable struct BasicCarPolicy <: Reinforce.AbstractPolicy end

msaroufim

# Environment setup
env = MountainCar()

function episode!(env, π = RandomPolicy())
  ep = Episode(env, π)
  for (s, a, r, s′) in ep
    gui(plot(env))
  end
  ep.total_reward, ep.niter
end

msaroufim

# Deterministic policy that is solving the problem
mutable struct BasicCarPolicy <: Reinforce.AbstractPolicy end

Reinforce.action(policy::BasicCarPolicy, r, s, A) = s.velocity < 0 ? 1 : 3

msaroufim

# Main part
R, n = episode!(env, BasicCarPolicy())
println("reward: $R, iter: $n")

msaroufim

# src/Reinforce.jl


abstract type AbstractPolicy end

"""
    a = action(policy, r, s, A)

Take in the last reward `r`, current state `s`,
and set of valid actions `A = actions(env, s)`,

msaroufim

mutable struct Episode{E<:AbstractEnvironment,P<:AbstractPolicy,F<:AbstractFloat}
  env::E
  policy::P
  total_reward::F # total reward of the episode
  last_reward::F
  niter::Int      # current step in this episode
  freq::Int       # number of steps between choosing actions
  maxn::Int       # max steps in an episode - should be constant during an episode
end

msaroufim

const min_position = -1.2
const max_position = 0.6
const max_speed = 0.07
const goal_position = 0.5
const min_start = -0.6
const max_start = 0.4

const car_width = 0.05
const car_height = car_width/2.0
const clearance = 0.2*car_height

msaroufim

const min_position = -1.2
const max_position = 0.6
const max_speed = 0.07
const goal_position = 0.5
const min_start = -0.6
const max_start = 0.4

const car_width = 0.05
const car_height = car_width/2.0
const clearance = 0.2*car_height

msaroufim

actions(env::MountainCar, s) = DiscreteSet(1:3)

msaroufim

finished(env::MountainCar, s′) = env.state.position >= goal_position