train_set = NonterminalFeaturesDataset(os.path.join(
args.data_dir, 'train'))
train_loader = DataLoader(train_set, batch_size=None)
val_set = NonterminalFeaturesDataset(os.path.join(args.data_dir, 'val'))
val_loader = DataLoader(train_set, batch_size=None)
gen_set = NonterminalFeaturesDataset(os.path.join(args.data_dir, 'gen'))
gen_loader = DataLoader(train_set, batch_size=None)
classifier = Classifier(args.features, args.nonterminals)
classifier = classifier.to(device)
optimizer = torch.optim.SGD(classifier.parameters(), lr=args.lr)
# Trainer and metrics
save_dict = {'classifier': classifier}
trainer = Engine(step_train(classifier, optimizer))
metric_names = ['loss', 'accuracy']
RunningAverage(Loss(F.cross_entropy, lambda x:
(x['y_pred'], x['y_true']))).attach(trainer, 'loss')
RunningAverage(Accuracy(lambda x: (x['y_pred'], x['y_true']))).attach(
trainer, 'accuracy')
# Evaluator and metrics
evaluator = Engine(step_train(classifier, None, train=False))
Accuracy(lambda x: (x['y_pred'], x['y_true'])).attach(
evaluator, 'accuracy')
# Begin training
run(args.run_name, save_dict, metric_names, trainer, evaluator,
train_loader, val_loader, gen_loader, args.epochs, 'accuracy')
DataSet=int(sys.argv[2]),
TestID=int(sys.argv[3]),
NewC=int(sys.argv[4]),
NetworkType=int(sys.argv[5]),
runs=int(sys.argv[6]),
Module=int(sys.argv[7]),
Mfit=boolArgv(sys.argv[8]),
FusionNType=int(sys.argv[9]),
FusionMfit=boolArgv(sys.argv[10]),
FLdata=boolArgv(sys.argv[11]),
FUSION=True)
elif len(sys.argv) == 8:
training.run(GPU_Device_ID=int(sys.argv[1]),
DataSet=int(sys.argv[2]),
ValidID=int(sys.argv[3]),
TestID=int(sys.argv[4]),
NetworkType=int(sys.argv[5]),
runs=int(sys.argv[6]),
Module=int(sys.argv[7]))
else:
print(
"argument errors, try\npython runfile.py <GPU_Device_ID> <DataSet> <ValidID> <TestID> <NetworkType> <runs> <Module>"
)
'''
Usage:
python Runscript.py 0 1 0 0 1 0
python Runscript.py 0 2 3 3 2 5
python Runscript.py 1 4 3 3 2 8
python Runscript.py 1 6 5 5 3 3
'''
import utils
import training
import testing
import sys
from config import Configuration
if __name__ == '__main__':
# Load condiguration from file or arguments
config = Configuration()
print("\nRun torcs env with this configuration?\n")
pprint(config.args)
# Create basic directories if not exists
utils.mkdir_resources(config.args['resources'])
while (True):
question = input(
"\n\nPlease, press one key:\n 1. TRAIN\n 2. TEST\n 0. EXIT\n")
question = int(question)
if (question == 1):
utils.to_json(config.args)
training.run()
elif (question == 2):
testing.run()
elif (question == 0):
sys.exit(0)
else:
continue
from get_train_config import training_configs
# -o D:\Projects\Myeloma_BM_mapping\Data\20200401\models -d D:\Projects\Myeloma_BM_mapping\Data\20200401\combined_polyscope-rectangles\patch -n 4
if __name__ == '__main__':
configs, args = training_configs()
if args.cluster:
print('*' * 50)
print('job started running on a cluster')
print('*' * 50)
print('*' * 50)
print('running configuration number: {}'.format(args.config_id))
print('*' * 50)
if args.config_id == '-1':
for n in range(1, len(configs) + 1):
print('*' * 50)
print('running configuration number: {}'.format(n))
print('*' * 50)
run(**configs[str(n)], all_params=configs[str(n)])
else:
run(**configs[args.config_id])
else:
for n in range(1, len(configs) + 1):
print('*' * 50)
print('running configuration number: {}'.format(n))
print('*' * 50)
run(**configs[str(n)], all_params=configs[str(n)])
default=None,
help="Override train batch size")
parser.add_argument("--lr",
type=float,
default=None,
help="Override train learning rate")
parser.add_argument("--ep",
type=int,
default=None,
help="Override number of epochs")
args = parser.parse_args()
assert args.config is not None
assert args.config.exists()
# Define configuration mutations if certain cmd args are defined
mutations = {}
if args.bs is not None:
mutations["train_batch_size"] = args.bs
if args.lr is not None:
mutations["learning_rate"] = args.lr
if args.ep is not None:
mutations["num_epochs"] = args.ep
# Pass configuration file into py_config_runner.ConfigObject
# and fetch configuration parameters as attributes
# see inside run() function
config = ConfigObject(args.config, mutations=mutations)
run(config)
model = SyntaxTreeNetwork(train_set.input_size, args.hidden_size,
train_set.num_nonterminal_rules,
train_set.num_nonterminals)
elif args.model == 'tree_network':
model = TreeNetwork(train_set.input_size, args.hidden_size)
elif args.model == 'rnn':
model = RNN(train_set.input_size, args.hidden_size)
else:
raise NotImplementedError('unknown model type {}'.format(args.model))
model = model.to(device)
classifier = Classifier(args.hidden_size, train_set.n_classes)
classifier = classifier.to(device)
optimizer = torch.optim.SGD(list(model.parameters()) +
list(classifier.parameters()),
lr=args.lr)
# Trainer and metrics
save_dict = {'model': model, 'classifier': classifier}
trainer = Engine(step_train(model, optimizer))
metric_names = ['loss', 'accuracy']
RunningAverage(Loss(F.cross_entropy, lambda x:
(x['y_pred'], x['y_true']))).attach(trainer, 'loss')
RunningAverage(Accuracy(lambda x: (x['y_pred'], x['y_true']))).attach(
trainer, 'accuracy')
# Begin training
run(args.run_name, save_dict, metric_names, trainer, None, train_loader,
None, None, args.epochs, 'accuracy')
logging.info('All data from raster will be kept in %s',
allDataPreProcessedPath)
# make 32x32 samples saves them at samplesPaths
# and create List of all samples [ name, classId, % of pixels of the class] and saves it to dataset_folder_path
samplesList = raster.sampleriseRaster(
mainRasterPath, samplesPaths,
dataset_folder_path + processName + "SamplesList", saveSamplesToDisc)
#if you skip line before uncoment below
# with open(dataset_folder_path + processName + "SamplesList.pkl", mode='rb') as file:
# samplesList = pickle.load(file, encoding='latin1')
# prepers input for CNN
dataConversion.prepBatches(samplesList, samplesPaths, 7000,
dataset_folder_path + processName +
"TrainingBatch") # for prepered set
dataConversion.prepBatches(samplesList, samplesPaths, 61168,
allDataPreProcessedPath + processName +
"AllDataBatch") # for All data set
# trains the network
training.run(dataset_folder_path, save_model_path,
processName + "TrainingBatch")
# test data evaluating
evalTrainings.test_model(save_model_path, dataset_folder_path)
# evaluate network on all data
processAllDataForValidation.preprocess_and_save_data(allDataPreProcessedPath +
processName)
evaluating.test_model(
save_model_path, allDataPreProcessedPath
) # allDataPreProcessedPath must be path to folder, all data '.p' will be used during evaluation
def loop():
switchApp("umamusume")
if not reg.has(windowTitle):
return False
if reg.has(
Pattern("1623306979891.png").similar(0.95).targetOffset(-226, 0)):
reg.click()
return True
if reg.has("1616475107941.png"):
reg.click()
return True
if reg.has(nextLabel):
reg.click()
return True
# not in main menu -> skip
if not reg.has(mainMenuHeader):
return True
if (reg.has("1615928744521.png")
or reg.has(Pattern("1626317018337.png").similar(0.85))
) and reg.has("1616367046236.png"):
reg.click("1616367046236.png")
waitClick(okLabel)
return True
if reg.has("1626393239326.png"):
reg.click()
if reg.has("1616366183098.png", 2):
waitClick(okLabel)
return True
else:
# ポップアップが出なかった場合、誤クリックなので流す
sleep(1)
if reg.has(raceDayLabel):
reg.click(raceDayLabel)
race.runRace(True)
try:
waitClick(nextLabel)
sleep(2)
waitClick(nextLabel)
return True
except FindFailed:
print("Final")
return True
isRaceDay = race.isRaceDay()
if (isRaceDay or objectiveRegion.has(notYetLabel)
) and not reg.has(raceKeyLabel) and reg.has(raceLabel):
reg.click(raceLabel)
if reg.has("1615978551852.png", 5):
reg.click("1615978573903.png")
sleep(2)
# returnせずに次の判定にうつる
else:
result = race.runRace(False)
if result:
return True
else:
# returnせず次の判定にうつる
sleep(2)
if reg.has(gageHalf):
if reg.has(restLabel):
reg.click(restLabel)
waitClick(okLabel)
return True
if reg.has("1616407518547.png"):
reg.click("1616407518547.png")
waitClick(okLabel)
return True
if reg.has(trainingLabel):
reg.click(trainingLabel)
reg.wait("1626318791352.png")
training.run([1, 0.6, 1, 0.3, 0.4])
return True
# 何にもヒットしなかったのでもう一度ループ
return True