alexmlamb

Goal:
Agent:(0,0).  Block:(4,4)
x        .       .       .       .       .
.        .       .       .       .       .
.        .       .       .       .       .
.        .       .       .       .       .
.        .       .       .       b       .
.        .       .       .       .       .
State at step 0:
Agent:(3,0).  Block:(1,3)

alexmlamb

Goal:
Agent:(4,3).  Block:(1,3)
.        .       .       .       .
.        .       .       .       .
.        .       .       .       .
.        b       .       .       x
.        .       .       .       .
State at step 0:
Agent:(3,2).  Block:(0,1)
.        .       .       .       .

alexmlamb

grtr_txt = 'Alex Lamb is here.  '
pred_txt = 'Alex Lamb is here.  '

def counts(string, size=3):
    d = {}
    for i in range(0,len(string)-size+1):
        sub = tuple(string[i:i+size])

        if not sub in d:

alexmlamb

gt_txt = 'Iamalexlambfirstnamealex'
pred_txt = 'Iamalexlambfir..stnamealex'

def counts(string, size=3):
    d = {}
    for i in range(0,len(string)-size+1):
        sub = string[i:i+size]

        if not sub in d:

alexmlamb

import numpy as np
import torch
import random
from torch.autograd import Variable, grad
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

alexmlamb

import random

def rollout(T):
    s = 0
    bought = 0
    sold = 0

    for i in range(T):
        if random.uniform(0,1) < 0.5:
            s += 1

alexmlamb

def gaussian_likelihood(mu, sigma, t):
  l = 0.5 * ((t - mu)/sigma)**2 + torch.log(sigma)
  return l

assert(state.shape[0] == action.shape[0])
assert(state.shape[0] == target_Q.shape[0])
ind = torch.randperm(state.shape[0])
state_rs = state[ind]
action_rs = action[ind]
target_Q_rs = target_Q[ind]

alexmlamb

assert(state.shape[0] == action.shape[0])
assert(state.shape[0] == target_Q.shape[0])
ind = torch.randperm(state.shape[0])
state_rs = state[ind]
action_rs = action[ind]
target_Q_rs = target_Q[ind]
def make_critic_loss(s,a,t):  
  current_Q1, current_Q2 = self.critic(s, a)
  return F.mse_loss(current_Q1, t) + F.mse_loss(current_Q2, t)
alpha = 0.1

alexmlamb

import matplotlib
matplotlib.use('Agg')
from matplotlib import pyplot as plt
from torch.utils.data import TensorDataset, DataLoader
from tqdm import tqdm_notebook as tqdm
import numpy as np
import torch
import random

torch.manual_seed(52)

alexmlamb

import torch
from torchvision.utils import save_image

from  torch.nn.modules.upsampling import Upsample
m = Upsample(scale_factor=84//4, mode='nearest')


'''
Image is: (bs, 3, 84, 84)