def trainNTM():
t = Tasks()
x_train, y_train = t.sequence_type_1(2000)
ntm = NTM(10, 20)
ntm.train(x_train, y_train, 1, maxEpoch=25, learning_rate=0.0006)
def trainNTM():
ntm = NTM(10, 14)
X, y = [], []
for i in range(10):
tempX, tempY = getData("data/observations_"+str(i*500)+".npy", "data/actions_"+str(i*500)+".npy")
X.extend(tempX)
y.extend(tempY)
print(len(X), len(y))
ntm.train(X, y, 1)
def trainNTM():
ntm = NTM(10, 14)
X, y = getData()
ntm.train(X, y, 1)
bit = 1
else:
sampleIdx = sampleIdxGreen
bit = 0
start = random.sample(sampleIdx, 1)[0]
sampleIdx.remove(start)
print(start)
imageSequence = np.load("sequences/image/imageSequence_" + str(start) +
".npy")
robotGpsSequence = np.load("sequences/robot/robotGpsSequence_" +
str(start) + ".npy")
actionSequence = np.load("sequences/action/actionSequence_" +
str(start) + ".npy")
imageSequence = torch.from_numpy(imageSequence).float()
robotGpsSequence = torch.from_numpy(robotGpsSequence).float()
y = torch.from_numpy(actionSequence).float()
loss = ntm.train(imageSequence, y, robotGpsSequence, learning_rate)
losses.append(loss.detach().numpy())
print(i, inEpochCtr, loss)
ctr += 1
torch.save(ntm.state_dict(), "ntm.pt")
np.save("losses/ntm", np.array(losses))
