def train(model, optim, steps, BS=128, gpu=False):
if gpu is True: [x.cuda_() for x in get_parameters([model, optim])]
losses, accuracies = [], []
for i in (t := trange(steps, disable=os.getenv('CI') is not None)):
samp = np.random.randint(0, X_train.shape[0], size=(BS))
x = Tensor(X_train[samp].reshape((-1, 28 * 28)).astype(np.float32),
gpu=gpu)
Y = Y_train[samp]
y = np.zeros((len(samp), 10), np.float32)
# correct loss for NLL, torch NLL loss returns one per row
y[range(y.shape[0]), Y] = -10.0
y = Tensor(y, gpu=gpu)
# network
out = model.forward(x)
# NLL loss function
loss = out.mul(y).mean()
optim.zero_grad()
loss.backward()
optim.step()
cat = np.argmax(out.cpu().data, axis=1)
accuracy = (cat == Y).mean()
# printing
loss = loss.cpu().data
losses.append(loss)
accuracies.append(accuracy)
t.set_description("loss %.2f accuracy %.2f" % (loss, accuracy))
def step_tinygrad(optim, kwargs={}, gpu=False):
net = TinyNet()
optim = optim([net.x, net.W], **kwargs)
if gpu is True: [x.cuda_() for x in get_parameters([net, optim])]
out = net.forward()
out.backward()
optim.step()
return net.x.cpu().data, net.W.cpu().data
def parameters(self):
return get_parameters(self)
# if you want to look at the outputs
"""
import matplotlib.pyplot as plt
plt.plot(out.data[0])
plt.show()
"""
return out, retimg
if __name__ == "__main__":
# instantiate my net
model = EfficientNet(int(os.getenv("NUM", "0")))
model.load_weights_from_torch()
if GPU:
[x.cuda_() for x in get_parameters(model)]
# category labels
import ast
lbls = fetch(
"https://gist.githubusercontent.com/yrevar/942d3a0ac09ec9e5eb3a/raw/238f720ff059c1f82f368259d1ca4ffa5dd8f9f5/imagenet1000_clsidx_to_labels.txt"
)
lbls = ast.literal_eval(lbls.decode('utf-8'))
# load image and preprocess
from PIL import Image
url = sys.argv[1]
if url == 'webcam':
import cv2
cap = cv2.VideoCapture(0)
cap.set(cv2.CAP_PROP_BUFFERSIZE, 1)
if __name__ == "__main__":
X_train, Y_train = load_cifar()
classes = 10
Tensor.default_gpu = os.getenv("GPU") is not None
TINY = os.getenv("TINY") is not None
TRANSFER = os.getenv("TRANSFER") is not None
if TINY:
model = TinyConvNet(classes)
elif TRANSFER:
model = EfficientNet(int(os.getenv("NUM", "0")), classes, has_se=True)
model.load_weights_from_torch()
else:
model = EfficientNet(int(os.getenv("NUM", "0")), classes, has_se=False)
parameters = get_parameters(model)
print("parameters", len(parameters))
optimizer = optim.Adam(parameters, lr=0.001)
#BS, steps = 16, 32
BS, steps = 64 if TINY else 16, 2048
for i in (t := trange(steps)):
samp = np.random.randint(0, X_train.shape[0], size=(BS))
img = X_train[samp].astype(np.float32)
st = time.time()
out = model.forward(Tensor(img))
fp_time = (time.time() - st) * 1000.0
