def transform_y(y_train):
# Transform y_train.
y_encoder = OneHotEncoder()
y_encoder.fit(y_train)
y_train = y_encoder.transform(y_train)
return y_train, y_encoder
if __name__ == '__main__':
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = np.squeeze(x_train.reshape((x_train.shape[0], -1)))
x_test = np.squeeze(x_test.reshape((x_test.shape[0], -1)))
y_train, y_encoder = transform_y(y_train)
y_test, _ = transform_y(y_test)
mlpModule = MlpModule(loss=classification_loss, metric=Accuracy, searcher_args={}, verbose=True)
# specify the fit args
data_transformer = DataTransformerMlp(x_train)
train_data = data_transformer.transform_train(x_train, y_train)
test_data = data_transformer.transform_test(x_test, y_test)
fit_args = {
"n_output_node": y_encoder.n_classes,
"input_shape": x_train.shape,
"train_data": train_data,
"test_data": test_data
}
mlpModule.fit(n_output_node=fit_args.get("n_output_node"),
input_shape=fit_args.get("input_shape"),
train_data=fit_args.get("train_data"),
test_data=fit_args.get("test_data"),
time_limit=24 * 60 * 60)
def transform_y(y_train):
# Transform y_train.
y_encoder = OneHotEncoder()
y_encoder.fit(y_train)
y_train = y_encoder.transform(y_train)
return y_train, y_encoder
if __name__ == '__main__':
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = np.squeeze(x_train.reshape((x_train.shape[0], -1)))
x_test = np.squeeze(x_test.reshape((x_test.shape[0], -1)))
y_train, y_encoder = transform_y(y_train)
y_test, _ = transform_y(y_test)
mlp_module = MlpModule(loss=classification_loss,
metric=Accuracy,
searcher_args={},
verbose=True)
# specify the fit args
data_transformer = DataTransformerMlp(x_train)
train_data = data_transformer.transform_train(x_train, y_train)
test_data = data_transformer.transform_test(x_test, y_test)
fit_args = {
"n_output_node": y_encoder.n_classes,
"input_shape": x_train.shape,
"train_data": train_data,
"test_data": test_data
}
mlp_module.fit(n_output_node=fit_args.get("n_output_node"),
input_shape=fit_args.get("input_shape"),
train_data=fit_args.get("train_data"),
test_data=fit_args.get("test_data"),
xtrains = np.zeros((m-1,n))
ytrains = np.zeros((m-1,n-2))
for i in range(n):
xtrains[:,i] = (2.0*xtrain[:,i]-(xmax[i]+xmin[i]))/(xmax[i]-xmin[i])
for i in range(n-2):
ytrains[:,i] = (2.0*ytrain[:,i]-(ymax[i]+ymin[i]))/(ymax[i]-ymin[i])
#%%
indices = np.random.randint(0,xtrains.shape[0],1000)
xtrainv = xtrains[indices]
ytrainv = ytrains[indices]
#%%
mlpModule = MlpModule(loss=regression_loss, metric=Accuracy, searcher_args={}, verbose=True)
data_transformer = DataTransformerMlp(xtrains)
train_data = data_transformer.transform_train(xtrains, ytrains)
test_data = data_transformer.transform_test(xtrainv, ytrainv)
fit_args = {
"n_output_node": ytrains.shape[1],
"input_shape": xtrains.shape,
"train_data": train_data,
"test_data": test_data
}
mlpModule.fit(n_output_node=fit_args.get("n_output_node"),
input_shape=fit_args.get("input_shape"),
train_data=fit_args.get("train_data"),
test_data=fit_args.get("test_data"),
time_limit=1 * 60* 60)
def train_autokeras(classes, alldata, labels, mtype, jsonfile, problemtype,
default_features):
## this is a CNN architecture
modelname = jsonfile[0:-5] + '_autokeras_%s' % (default_features)
TEST_FOLDER = modelname
x_train, x_test, y_train, y_test = train_test_split(alldata,
labels,
train_size=0.750,
test_size=0.250)
# we have to do some odd re-shapes to get the data loader to work for the autokeras module (keep this in mind when loading new data in)
x_train = x_train.reshape(x_train.shape + (1, ))
y_train = y_train.reshape(y_train.shape + (1, ) + (1, ))
x_test = x_test.reshape(x_test.shape + (1, ))
y_test = y_test.reshape(y_test.shape + (1, ) + (1, ))
print(x_train.shape)
print(y_train.shape)
tensor_x = torch.stack([torch.Tensor(i)
for i in x_train]) # transform to torch tensors
tensor_y = torch.stack([torch.Tensor(i) for i in y_train])
my_dataset = utils.TensorDataset(tensor_x, tensor_y) # create your datset
training_data = utils.DataLoader(my_dataset) # create your dataloader
tensor_x = torch.stack([torch.Tensor(i)
for i in x_test]) # transform to torch tensors
tensor_y = torch.stack([torch.Tensor(i) for i in y_test])
my_dataset = utils.TensorDataset(tensor_x, tensor_y) # create your datset
test_data = utils.DataLoader(my_dataset) # create your dataloader
print(test_data)
input_shape = x_train[0].shape
n_output_node = 1
# cnnModule = CnnModule(loss=classification_loss, metric=Accuracy, searcher_args={}, path=TEST_FOLDER, verbose=False)
if mtype == 'c':
# metric = Accuracy is for classification
# loss = classiciation_loss for classification
mlpModule = MlpModule(loss=classification_loss,
metric=Accuracy,
searcher_args={},
path=TEST_FOLDER,
verbose=True)
elif mtype == 'r':
# metric = MSE for regression
# loss = regression_loss for regression
mlpModule = MlpModule(loss=regression_loss,
metric=MSE,
searcher_args={},
path=TEST_FOLDER,
verbose=True)
timelimit = 60
print('training MLP model for %s hours' % (timelimit / (60 * 60)))
mlpModule.fit(n_output_node,
input_shape,
training_data,
test_data,
time_limit=timelimit)
mlpModule.final_fit(training_data,
test_data,
trainer_args=None,
retrain=False)
# # serialize model to JSON
# mlpModule.export_autokeras_model(modelname+'.pickle')
# print("\n Saved %s.pickle model to disk"%(modelname))
# # test opening model and making predictions
# model=pickle_from_file(modelname+'.pickle')
# results=model.evaluate(x_test, y_test)
# print(results)
cur_dir2 = os.getcwd()
try:
os.chdir(problemtype + '_models')
except:
os.mkdir(problemtype + '_models')
os.chdir(problemtype + '_models')
# now move all the files over to proper model directory
shutil.copytree(cur_dir2 + '/' + TEST_FOLDER,
os.getcwd() + '/' + TEST_FOLDER)
shutil.rmtree(cur_dir2 + '/' + TEST_FOLDER)
import numpy as np
from autokeras import MlpModule
from autokeras.nn.loss_function import classification_loss
from autokeras.nn.metric import Accuracy
X_train = np.array([[0, 0], [0, 1], [1, 0], [1, 1]]) # 异或训练集
y_train = np.array([0, 1, 1, 0])
X_test = np.array([[0, 0], [0, 1], [1, 0], [1, 1]]) # 异或训练集
y_test = np.array([0, 1, 1, 0])
mlpModule = MlpModule(loss=classification_loss,
metric=Accuracy,
searcher_args={},
verbose=True) # 多层神经网络模型
mlpModule.fit(
n_output_node=2, # 输出分几类
input_shape=(4, 2), # 输入的形状
train_data=X_train, # 训练集
test_data=y_train, # 验证集
time_limit=24 * 60 * 60) # 超时时间的设置