po-hsun-su / dprl Goto Github PK

dqn.lua:89: internal error in __mul: no metatable
stack traceback:
    [C]: in function '__mul'
    ...berwebid/Documents/_Development/AI learning/dprl/dqn.lua:89: in function 'setTarget'
    ...berwebid/Documents/_Development/AI learning/dprl/dqn.lua:103: in function 'learn'

require 'nn'
require 'optim'
local classic = require 'classic'
local memory = require 'dprl/memory'
local dqn = require 'dprl/dqn'

local numStates = 16
local numActions = 5
local qnet = nn.Sequential()
:add(nn.Linear(numStates,numActions))
:add(nn.Tanh())

local dqn_param = {replaySize = 32, batchSize = 4, discount = 0.9, epsilon = 0.1}
local optimConfig = {learningRate = 0.01, momentum = 0.0}
local optimMethod = optim.rmsprop
local dqnet = dqn(qnet, dqn_param, optimMethod, optimConfig)

--dummy data
local state = torch.Tensor(16):zero()
local action = dqnet:act(state)
local reward = torch.Tensor(1):fill(1)
local nextState = torch.Tensor(16):fill(1)
local terminal = torch.Tensor(1):zero()

local trans = {s = state, a = action:clone(), r = reward, ns = nextState, t = terminal}
local sampleTrans = dqnet:replay(trans)
dqnet:learn(sampleTrans)