args = parser.parse_args()
ratio = args.ratio
epoch = args.epoch
nBatch = args.nBatch
lr = args.lr
pPerEpoch = args.pPerEpoch if args.pDetail else -1
x_train, _, y_value, y_pred, x_test, y_test, traget_pred = datasets.getBigML(
args.data, ratio)
layer = layers.getLayer(args.data)
#control model
if args.pDetail:
print("\n======================== control model ========================")
control = pack.classifyModel(layer, optimArgs={"lr": lr})
control.train(x_train,
y_pred,
epoch,
nBatch,
printPerEpoch=pPerEpoch,
yType="long",
yTo2D=False,
printData=args.pDetail,
printAcc=args.pDetail)
control.save(getAbsPath("model/BigML_" + args.data + "_control.model"))
#copy model
if args.pDetail:
print("\n======================== copy model ========================")
copy = pack.classifyModel(layer, loss_func=args.loss, optimArgs={"lr": lr})
import datasets
import packagedModel as pack
x_train, y_train = datasets.getTrainedIris()
# layers = [("Linear", (5,5)), \
# ("ReLU", ()), \
# ("Linear", (5,5)), \
# ("ReLU", ()), \
# ("Linear", (5,3)), \
# ("Softmax", 1)]
layers = [("Linear", (5,2)), \
("ReLU", ()), \
("Linear", (2,3)), \
("Softmax", 1)]
# layers = [("Linear", (5,3)), \
# ("Softmax", 1)]
model = pack.classifyModel(layers, loss_func="BCELoss", optimArgs = {"lr" : 1e-3} )
model.train(x_train, y_train, epoch=1000, batch = len(x_train)//2, printPerEpoch=50)
model.save("iris.model")
for i in range(len(x_train)) :
predict = model.predict(x_train[i])
x_result = predict.argmax()
print(predict.argmax(), end=" ")
print(y_train[i].argmax())
import sys
sys.path.insert(1,
os.path.join(os.path.dirname(os.path.abspath(__file__)), ".."))
import packagedModel as pack
import datasets
import numpy as np
x_train, y_train, x_test, y_test = datasets.getMnist()
layers = [("Linear", (784,100)), \
("ReLU", ()), \
("Linear", (100,100)), \
("ReLU", ()), \
("Linear", (100,10)), \
("Softmax", 1)]
epoch = 10
batch = 100
if __name__ == '__main__':
model = pack.classifyModel(layers,
loss_func="CrossEntropyLoss",
optimArgs={"lr": 5e-5})
model.train(x_train, y_train, epoch, batch)
model.save(
os.path.join(os.path.dirname(os.path.abspath(__file__)),
"mnist.model"))
print(model.getAccuracy(x_test, y_test))
x_train, y_train, x_test, y_test = datasets.getMnist()
mask = np.random.choice(x_train.shape[0], nTrain, replace=False)
x_train = x_train[mask]
y_train = y_train[mask]
y_value = pre_model.predict(x_train)
layers = [("Linear", (784,100)), \
("ReLU", ()), \
("Linear", (100,100)), \
("ReLU", ()), \
("Linear", (100,10)), \
("Softmax", 1)]
#Origin model
print("\n======================== origin model ========================")
origin = pack.classifyModel(layers, optimArgs={"lr": 5e-5})
origin.train(x_train,
y_train,
epoch,
nBatch,
printPerEpoch=pPerEpoch,
yType="long",
yTo2D=False)
origin.save(
os.path.join(os.path.dirname(os.path.abspath(__file__)),
"model/origin_mnist.model"))
#copy model
print("\n======================== copy model ========================")
copy = pack.classifyModel(layers, loss_func="BCELoss", optimArgs={"lr": 5e-5})
copy.train(x_train, y_value, epoch, nBatch, printPerEpoch=pPerEpoch)
def getMnistModel():
import pretrain_mnist as mnist
result = pack.classifyModel(mnist.layers)
result.load(getAbsPath("pre_model/mnist.model"))
return result