diff --git a/temp_util/parse_ds_cn_arch.py b/temp_util/parse_ds_cn_arch.py
index e783b7da25414778e52fa16d09eff75cf944593c..76d33c195ca53a6e05d9d4d144788cd2f691cfc9 100644
--- a/temp_util/parse_ds_cn_arch.py
+++ b/temp_util/parse_ds_cn_arch.py
@@ -1,28 +1,38 @@
from ..MetaAugment.child_networks import *
from ..MetaAugment.main import create_toy, train_child_network
import torch
+import torchvision
import torchvision.datasets as datasets
import pickle
-def parse_ds_cn_arch(ds, ds_name, IsLeNet, transform):
- # open data and apply these transformations
+def parse_ds_cn_arch(ds, ds_name, IsLeNet):
if ds == "MNIST":
- train_dataset = datasets.MNIST(root='./datasets/mnist/train', train=True, download=True, transform=transform)
- test_dataset = datasets.MNIST(root='./datasets/mnist/test', train=False, download=True, transform=transform)
+ train_dataset = datasets.MNIST(root='./datasets/mnist/train',
+ train=True, download=True, transform=None)
+ test_dataset = datasets.MNIST(root='./datasets/mnist/test', train=False,
+ download=True, transform=torchvision.transforms.ToTensor())
elif ds == "KMNIST":
- train_dataset = datasets.KMNIST(root='./datasets/kmnist/train', train=True, download=True, transform=transform)
- test_dataset = datasets.KMNIST(root='./datasets/kmnist/test', train=False, download=True, transform=transform)
+ train_dataset = datasets.KMNIST(root='./datasets/kmnist/train',
+ train=True, download=True, transform=None)
+ test_dataset = datasets.KMNIST(root='./datasets/kmnist/test', train=False,
+ download=True, transform=torchvision.transforms.ToTensor())
elif ds == "FashionMNIST":
- train_dataset = datasets.FashionMNIST(root='./datasets/fashionmnist/train', train=True, download=True, transform=transform)
- test_dataset = datasets.FashionMNIST(root='./datasets/fashionmnist/test', train=False, download=True, transform=transform)
+ train_dataset = datasets.FashionMNIST(root='./datasets/fashionmnist/train',
+ train=True, download=True, transform=None)
+ test_dataset = datasets.FashionMNIST(root='./datasets/fashionmnist/test', train=False,
+ download=True, transform=torchvision.transforms.ToTensor())
elif ds == "CIFAR10":
- train_dataset = datasets.CIFAR10(root='./datasets/cifar10/train', train=True, download=True, transform=transform)
- test_dataset = datasets.CIFAR10(root='./datasets/cifar10/test', train=False, download=True, transform=transform)
+ train_dataset = datasets.CIFAR10(root='./datasets/cifar10/train',
+ train=True, download=True, transform=None)
+ test_dataset = datasets.CIFAR10(root='./datasets/cifar10/test', train=False,
+ download=True, transform=torchvision.transforms.ToTensor())
elif ds == "CIFAR100":
- train_dataset = datasets.CIFAR100(root='./datasets/cifar100/train', train=True, download=True, transform=transform)
- test_dataset = datasets.CIFAR100(root='./datasets/cifar100/test', train=False, download=True, transform=transform)
+ train_dataset = datasets.CIFAR100(root='./datasets/cifar100/train',
+ train=True, download=True, transform=None)
+ test_dataset = datasets.CIFAR100(root='./datasets/cifar100/test', train=False,
+ download=True, transform=torchvision.transforms.ToTensor())
elif ds == 'Other':
- dataset = datasets.ImageFolder('./datasets/upload_dataset/'+ ds_name, transform=transform)
+ dataset = datasets.ImageFolder('./datasets/upload_dataset/'+ ds_name, transform=None)
len_train = int(0.8*len(dataset))
train_dataset, test_dataset = torch.utils.data.random_split(dataset, [len_train, len(dataset)-len_train])
@@ -43,12 +53,12 @@ def parse_ds_cn_arch(ds, ds_name, IsLeNet, transform):
if IsLeNet == "LeNet":
- model = LeNet(img_height, img_width, num_labels, img_channels)
+ child_architecture = LeNet(img_height, img_width, num_labels, img_channels)
elif IsLeNet == "EasyNet":
- model = EasyNet(img_height, img_width, num_labels, img_channels)
+ child_architecture = EasyNet(img_height, img_width, num_labels, img_channels)
elif IsLeNet == 'SimpleNet':
- model = SimpleNet(img_height, img_width, num_labels, img_channels)
+ child_architecture = SimpleNet(img_height, img_width, num_labels, img_channels)
else:
- model = pickle.load(open(f'datasets/childnetwork', "rb"))
+ child_architecture = pickle.load(open(f'datasets/childnetwork', "rb"))
- return train_dataset, test_dataset, model
\ No newline at end of file
+ return train_dataset, test_dataset, child_architecture
\ No newline at end of file
diff --git a/temp_util/wapp_util.py b/temp_util/wapp_util.py
index 5ea55018b78337fd71951d4d1d0ce3847ff50ad8..cde572f7f1e0ba1590fb5685b7212f4d0b3b173a 100644
--- a/temp_util/wapp_util.py
+++ b/temp_util/wapp_util.py
@@ -15,46 +15,62 @@ import MetaAugment.child_networks as cn
from MetaAugment.main import create_toy
import pickle
-
+from pprint import pprint
+from .parse_ds_cn_arch import parse_ds_cn_arch
def parse_users_learner_spec(
+ # things we need to feed into string parser
+ ds,
+ ds_name,
+ IsLeNet,
# aalearner type
auto_aug_learner,
# search space settings
- ds,
- ds_name,
exclude_method,
num_funcs,
num_policies,
num_sub_policies,
# child network settings
toy_size,
- IsLeNet,
batch_size,
early_stop_num,
iterations,
learning_rate,
max_epochs
):
+ train_dataset, test_dataset, child_archi = parse_ds_cn_arch(
+ ds,
+ ds_name,
+ IsLeNet
+ )
"""
The website receives user inputs on what they want the aa_learner
to be. We take those hyperparameters and return an aa_learner
"""
if auto_aug_learner == 'UCB':
- policies = aal.ucb_learner.generate_policies(num_policies, num_sub_policies)
- q_values, best_q_values = aal.ucb_learner.run_UCB1(
- policies,
- batch_size,
- learning_rate,
- ds,
- toy_size,
- max_epochs,
- early_stop_num,
- iterations,
- IsLeNet,
- ds_name
- )
- best_q_values = np.array(best_q_values)
+ learner = aal.ucb_learner(
+ # parameters that define the search space
+ sp_num=num_sub_policies,
+ p_bins=11,
+ m_bins=10,
+ discrete_p_m=True,
+ # hyperparameters for when training the child_network
+ batch_size=batch_size,
+ toy_size=toy_size,
+ learning_rate=learning_rate,
+ max_epochs=max_epochs,
+ early_stop_num=early_stop_num,
+ # ucb_learner specific hyperparameter
+ num_policies=num_policies
+ )
+ pprint(learner.policies)
+
+ learner.learn(
+ train_dataset=train_dataset,
+ test_dataset=test_dataset,
+ child_network_architecture=child_archi,
+ iterations=5
+ )
elif auto_aug_learner == 'Evolutionary Learner':
network = cont_n.evo_controller(fun_num=num_funcs, p_bins=1, m_bins=1, sub_num_pol=1)
child_network = cn.LeNet()
@@ -71,68 +87,28 @@ def parse_users_learner_spec(
)
learner.run_instance()
elif auto_aug_learner == 'Random Searcher':
- # As opposed to when ucb==True, `ds` and `IsLenet` are processed outside of the agent
- # This system makes more sense for the user who is not using the webapp and is instead
- # using the library within their code
- download = True
- if ds == "MNIST":
- train_dataset = datasets.MNIST(root='./MetaAugment/datasets/mnist/train', train=True, download=download)
- test_dataset = datasets.MNIST(root='./MetaAugment/datasets/mnist/test', train=False,
- download=download, transform=torchvision.transforms.ToTensor())
- elif ds == "KMNIST":
- train_dataset = datasets.KMNIST(root='./MetaAugment/datasets/kmnist/train', train=True, download=download)
- test_dataset = datasets.KMNIST(root='./MetaAugment/datasets/kmnist/test', train=False,
- download=download, transform=torchvision.transforms.ToTensor())
- elif ds == "FashionMNIST":
- train_dataset = datasets.FashionMNIST(root='./MetaAugment/datasets/fashionmnist/train', train=True, download=download)
- test_dataset = datasets.FashionMNIST(root='./MetaAugment/datasets/fashionmnist/test', train=False,
- download=download, transform=torchvision.transforms.ToTensor())
- elif ds == "CIFAR10":
- train_dataset = datasets.CIFAR10(root='./MetaAugment/datasets/cifar10/train', train=True, download=download)
- test_dataset = datasets.CIFAR10(root='./MetaAugment/datasets/cifar10/test', train=False,
- download=download, transform=torchvision.transforms.ToTensor())
- elif ds == "CIFAR100":
- train_dataset = datasets.CIFAR100(root='./MetaAugment/datasets/cifar100/train', train=True, download=download)
- test_dataset = datasets.CIFAR100(root='./MetaAugment/datasets/cifar100/test', train=False,
- download=download, transform=torchvision.transforms.ToTensor())
- elif ds == 'Other':
- dataset = datasets.ImageFolder('./MetaAugment/datasets/'+ ds_name)
- len_train = int(0.8*len(dataset))
- train_dataset, test_dataset = torch.utils.data.random_split(dataset, [len_train, len(dataset)-len_train])
-
- # check sizes of images
- img_height = len(train_dataset[0][0][0])
- img_width = len(train_dataset[0][0][0][0])
- img_channels = len(train_dataset[0][0])
- # check output labels
- if ds == 'Other':
- num_labels = len(dataset.class_to_idx)
- elif ds == "CIFAR10" or ds == "CIFAR100":
- num_labels = (max(train_dataset.targets) - min(train_dataset.targets) + 1)
- else:
- num_labels = (max(train_dataset.targets) - min(train_dataset.targets) + 1).item()
- # create toy dataset from above uploaded data
- train_loader, test_loader = create_toy(train_dataset, test_dataset, batch_size, toy_size)
- # create model
- if IsLeNet == "LeNet":
- model = cn.LeNet(img_height, img_width, num_labels, img_channels)
- elif IsLeNet == "EasyNet":
- model = cn.EasyNet(img_height, img_width, num_labels, img_channels)
- elif IsLeNet == 'SimpleNet':
- model = cn.SimpleNet(img_height, img_width, num_labels, img_channels)
- else:
- model = pickle.load(open(f'datasets/childnetwork', "rb"))
-
- # use an aa_learner. in this case, a rs learner
- agent = aal.randomsearch_learner(batch_size=batch_size,
- learning_rate=learning_rate,
- toy_size=toy_size,
- max_epochs=max_epochs,
- early_stop_num=early_stop_num,
- )
+ agent = aal.randomsearch_learner(
+ sp_num=num_sub_policies,
+ batch_size=batch_size,
+ learning_rate=learning_rate,
+ toy_size=toy_size,
+ max_epochs=max_epochs,
+ early_stop_num=early_stop_num,
+ )
+ agent.learn(train_dataset,
+ test_dataset,
+ child_network_architecture=child_archi,
+ iterations=iterations)
+ elif auto_aug_learner == 'GRU Learner':
+ agent = aal.gru_learner(
+ sp_num=num_sub_policies,
+ batch_size=batch_size,
+ learning_rate=learning_rate,
+ toy_size=toy_size,
+ max_epochs=max_epochs,
+ early_stop_num=early_stop_num,
+ )
agent.learn(train_dataset,
- test_dataset,
- child_network_architecture=model,
- iterations=iterations)
- elif auto_aug_learner == 'Genetic Learner':
- pass
\ No newline at end of file
+ test_dataset,
+ child_network_architecture=child_archi,
+ iterations=iterations)
\ No newline at end of file