def makeNewAreaDataset():
new_data = rawdata.RawData.load(locNames='untrain',
special_layers='all',
new_data='not_none')
newDataSet = dataset.Dataset(new_data, dataset.Dataset.vulnerablePixels)
pointLst = newDataSet.toListTest(newDataSet.points)
# ptList = masterDataSet.sample(sampleEvenly=False)
# pointLst = random.sample(pointLst, SAMPLE_SIZE)
test = dataset.Dataset(new_data, pointLst)
return test
def openDatasets():
data = rawdata.load()
masterDataSet = dataset.Dataset(data, dataset.Dataset.vulnerablePixels)
ptList = masterDataSet.sample(sampleEvenly=False)
# masterDataSet.points = dataset.Dataset.toDict(ptList)
trainPts, validatePts, testPts = util.partition(ptList, ratios=[.6,.7])
train = dataset.Dataset(data, trainPts)
validate = dataset.Dataset(data, validatePts)
test = dataset.Dataset(data, testPts)
return train, validate, test
def openDatasets():
data = rawdata.RawData.load(locNames='all', special_layers='all')
masterDataSet = dataset.Dataset(
data, dataset.Dataset.vulnerablePixels
) #this loops through vulnerablePixels for each location... should grab all veg image
sample_size = 100
print("SAMPLE SIZE: ", sample_size)
ptList = masterDataSet.sample(goalNumber=sample_size,
sampleEvenly=False) #goalNumber=sample_size,
trainPts, validatePts, testPts = util.partition(ptList,
ratios=[.7, .8]) #.85,.99
train = dataset.Dataset(data, trainPts)
validate = dataset.Dataset(data, validatePts)
test = dataset.Dataset(data, testPts)
return train, validate, test
def makeSmallDatasets(pass_arr):
data = pass_arr[0]
set_type = pass_arr[1]
type_str = pass_arr[2]
return_dict = {}
return_dict[type_str] = dataset.Dataset(data, set_type)
return return_dict
def get_data(batch_size, data_name, data_root='./my_ai/'):
data_loader = data.DataLoader(
dataset.Dataset(
path=data_root,
transform_data=transforms.Compose([
# transforms.RandomHorizontalFlip(),
# channel_change(),
color_change(),
ToTensor(),
tensor_pad(28)
]),
transform_labels=transforms.Compose([
# transforms.RandomHorizontalFlip(),
ToTensor(),
tensor_pad(28)
]),
data_name=data_name
),
batch_size=batch_size,
shuffle=True,
num_workers=1
)
return data_loader
def test():
ds = dataset.Dataset()
# ds2 = ds.copy()
# ds.save2()
for i in ds.getDays():
print(i)
print(ds)
pass
def setup_data_generators(
metadata,
folder_images,
field_class="dx",
test_size=0.1,
validation_size=0.2,
aux_data=False,
augment=True,
batch_size=50,
balancing=True,
seed=None) -> Tuple[dataset.Dataset, dataset.Dataset, dataset.Dataset]:
if seed is None:
seed = np.random.randint(0, 1e7)
train_data, test_data, _, _ = model_selection.train_test_split(
metadata,
metadata[field_class],
test_size=0.1,
stratify=metadata[field_class],
random_state=seed)
train_data, validate_data, _, _ = model_selection.train_test_split(
train_data,
train_data[field_class],
test_size=0.2,
stratify=train_data[field_class],
random_state=seed)
# set up image generators
get_dataset = (lambda data, aux_data, class_order=None, augment=True:
dataset.Dataset(folder_images,
data,
target_size=(300, 225),
augmentation=augment,
aux_data=aux_data,
batch_size=batch_size,
class_order=class_order))
# sync class order with training generator
train_gen = get_dataset(train_data, False)
test_gen = get_dataset(test_data,
False,
class_order=train_gen.unique_classes,
augment=False)
validate_gen = get_dataset(validate_data,
False,
class_order=train_gen.unique_classes)
# balance datasets
if balancing:
train_gen.balance(mode="upsampling", aggressiveness=0.7)
test_gen.balance(mode="upsampling", aggressiveness=0.7)
validate_gen.balance(mode="upsampling", aggressiveness=0.7)
return train_gen, test_gen, validate_gen
def predict(self):
selectedBurns = []
mod = self.burnList.model()
for index in range(mod.rowCount()):
i = mod.item(index)
# print(i.checkState())
if i.checkState() == QtCore.Qt.Checked:
selectedBurns.append(i.text())
print('opening the data for the burns,', selectedBurns)
data = rawdata.RawData.load(burnNames=selectedBurns, dates='all')
ds = dataset.Dataset(data, dataset.Dataset.vulnerablePixels)
from lib import model
modelFileName = self.modelLineEdit.text()
print('loading model', modelFileName)
mod = model.load(modelFileName)
print(mod)
