class DatasetRealSmolBulletV2NoSim(Dataset): def __init__(self, path="~/data/sim2real/data-realigned-v3-{}-bullet.npz", train=True): super().__init__() ds_name = "train" if not train: ds_name = "test" path_ext = path.format(ds_name) print("using dataset:", path_ext) self.ds = DatasetProduction() self.ds.load(path_ext) def __len__(self): return len(self.ds.current_real) def format_data(self, idx): return (np.hstack((self.ds.current_real[idx], self.ds.action[idx])), self.ds.next_real[idx] - self.ds.current_real[idx]) def __getitem__(self, idx): x, y = self.format_data(idx) return {"x": x, "y": y}
def __init__(self, path="~/data/sim2real/data-realigned-v3-{}-bullet.npz", train=True): super().__init__() ds_name = "train" if not train: ds_name = "test" path_ext = path.format(ds_name) print("using dataset:", path_ext) self.ds = DatasetProduction() self.ds.load(path_ext)
import time import numpy as np import os import matplotlib.pyplot as plt import gym import gym_ergojr import torch from gym_ergojr.sim.single_robot import SingleRobot from hyperdash import Experiment from s2rr.movements.dataset import DatasetProduction from torch.autograd import Variable from simple_joints_lstm.lstm_net_real_v3 import LstmNetRealv3 import torch.nn.functional as F ds = DatasetProduction() ds.load("~/data/sim2real/data-realigned-v3-{}-bullet.npz".format("train")) net = LstmNetRealv3(nodes=128, layers=3) if torch.cuda.is_available(): net = net.cuda() HIDDEN_NODES = 128 LSTM_LAYERS = 3 EXPERIMENT = 1 EPOCHS = 5 MODEL_PATH = "./trained_models/lstm_real_vX5_exp{}_l{}_n{}.pt".format( EXPERIMENT, LSTM_LAYERS, HIDDEN_NODES) def double_unsqueeze(data): return torch.unsqueeze(torch.unsqueeze(data, dim=0), dim=0)