def test_rmsprop(self):
np.random.seed(1337)
model = TinyBobNet()
optimizer = optim.RMSprop(model.parameters(), lr=0.0002)
train(model,
X_train,
Y_train,
optimizer,
steps=1000,
device=self.device)
assert evaluate(model, X_test, Y_test, device=self.device) > 0.95
def test_conv(self):
np.random.seed(1337)
model = TinyConvNet()
optimizer = optim.Adam(model.parameters(), lr=0.001)
train(model,
X_train,
Y_train,
optimizer,
steps=200,
device=self.device)
assert evaluate(model, X_test, Y_test, device=self.device) > 0.95
def test_sgd(self): np.random.seed(1337) model = TinyBobNet() optimizer = optim.SGD(model.parameters(), lr=0.001) train(model, X_train, Y_train, optimizer, steps=1000) assert evaluate(model, X_test, Y_test) > 0.95
lmbd = 0.00025
lossfn = lambda out,y: out.mul(y).mean() + lmbd*(model.weight1.abs() + model.weight2.abs()).sum()
X_train, Y_train, X_test, Y_test = fetch_mnist()
steps = len(X_train)//BS
np.random.seed(1337)
if QUICK:
steps = 1
X_test, Y_test = X_test[:BS], Y_test[:BS]
model = BigConvNet()
if len(sys.argv) > 1:
try:
model.load(sys.argv[1])
print('Loaded weights "'+sys.argv[1]+'", evaluating...')
evaluate(model, X_test, Y_test, BS=BS)
except:
print('could not load weights "'+sys.argv[1]+'".')
if GPU:
params = get_parameters(model)
[x.cuda_() for x in params]
for lr, epochs in zip(lrs, epochss):
optimizer = optim.Adam(model.parameters(), lr=lr)
for epoch in range(1,epochs+1):
#first epoch without augmentation
X_aug = X_train if epoch == 1 else augment_img(X_train)
train(model, X_aug, Y_train, optimizer, steps=steps, lossfn=lossfn, gpu=GPU, BS=BS)
accuracy = evaluate(model, X_test, Y_test, BS=BS)
model.save('examples/checkpoint'+str("%.0f" % (accuracy*1.0e6)))
return ds_X_train, ds_Y_train, ds_X_test, ds_Y_test
from tinygrad.optim import Adam
if __name__ == "__main__":
model = Transformer(10, 6, 2, 128, 4)
X_train, Y_train, X_test, Y_test = make_dataset()
lr = 0.003
for i in range(10):
optim = Adam(get_parameters(model), lr=lr)
train(model, X_train, Y_train, optim, 50, BS=64)
acc, Y_test_preds = evaluate(model,
X_test,
Y_test,
num_classes=10,
return_predict=True)
lr /= 1.2
print(f'reducing lr to {lr:.4f}')
if acc > 0.998:
wrong = 0
for k in range(len(Y_test_preds)):
if (Y_test_preds[k] != Y_test[k]).any():
wrong += 1
a, b, c, x = X_test[k, :2], X_test[k, 2:4], Y_test[
k, -3:], Y_test_preds[k, -3:]
print(
f'{a[0]}{a[1]} + {b[0]}{b[1]} = {x[0]}{x[1]}{x[2]} (correct: {c[0]}{c[1]}{c[2]})'
)
print(f'Wrong predictions: {wrong}, acc = {acc:.4f}')
ds.append([
i // 10, i % 10, j // 10, j % 10, s // 100, (s // 10) % 10,
s % 10
])
random.shuffle(ds)
ds = np.array(ds)
ds_X = ds[:, 0:6]
ds_Y = np.copy(ds[:, 1:])
ds_X_train, ds_X_test = ds_X[0:8000], ds_X[8000:]
ds_Y_train, ds_Y_test = ds_Y[0:8000], ds_Y[8000:]
return ds_X_train, ds_Y_train, ds_X_test, ds_Y_test
from tinygrad.optim import Adam
if __name__ == "__main__":
model = Transformer(10, 6, 2, 128, 4)
X_train, Y_train, X_test, Y_test = make_dataset()
optim = Adam(get_parameters(model), lr=0.001)
for i in range(5):
train(model,
X_train,
Y_train,
optim,
500,
BS=32,
device=Device.GPU if os.getenv("GPU") else Device.CPU)
evaluate(model, X_test, Y_test, num_classes=10)
