def run(output, subjects, ae_weights, clusters):
# subjects = Subjects.from_data(path)
fname = subjects
with open(subjects, 'rb') as file:
subjects = pickle.load(file)
logger.info('Done loading subjects')
config = Config(**{
'n_clusters': int(clusters or 10),
'batch_size': 256,
'lr': 0.01,
'momentum': 0.9,
'tol': 0.001,
'maxiter': 2e4,
'update_interval': 140,
'save_dir': output,
'ae_weights': ae_weights or os.path.join(output, 'ae_weights.h5'),
'subjects': fname,
})
config.dump()
cluster = Cluster.create(subjects, config)
logger.info('Training model')
cluster.train()
pred = cluster.predictions
logger.info('Done training network')
# if save:
# pickle.dump(cluster, open(save, 'wb'))
interact(locals())
def load(config):
config = Config.load(config)
subjects = pickle.load(open(config.subjects, 'rb'))
cluster = Cluster.create(subjects, config)
logger.info('Training model')
cluster.train()
pred = cluster.predictions
logger.info('Done training network')
interact(locals())
def main(path, name, lr):
config = Config.load(_config)
print(config.__dict__)
subjects = pickle.load(open(config.subjects, 'rb'))
labels = pe.Aggregate.load('mh2')
truth = pe.Aggregate.load('mh2_gold')
_labels = labels.subject_labels(), truth.subject_labels()
subjects = subjects.subset(truth.labeled_subjects())
truth.apply_labels(subjects)
cluster = Cluster.create(subjects, config)
cluster.initialize()
mapping = DEC_Updater(cluster, *_labels, train, validate, 0.95)
print(mapping.scores)
optimizer = optimizers.SGD(lr=float(lr))
mapping.init_model('categorical_crossentropy', optimizer)
print(mapping.scores)
weights_fname = os.path.join(path, name)
kwargs = {
'epochs':
500,
'batch_size':
256,
'callbacks': [
ModelCheckpoint(weights_fname,
monitor='val_loss',
save_best_only=True,
save_weights_only=True,
mode='min'),
EarlyStopping(monitor='val_loss',
min_delta=0,
patience=20,
verbose=0,
mode='min')
]
}
def fit(model, train, val):
model.fit(*train, validation_data=val, **kwargs)
# def xy(train, subjects):
# return subjects.get_xy_volunteer(labels.data['subjects'], False)
mapping.apply_mapping(fit)
print(mapping.cluster_mapping())
print('scores:', utils.pd_scores(mapping.scores))
def test(config, new):
config = Config.load(config)
subjects = pickle.load(open(config.subjects, 'rb'))
cluster = Cluster.create(subjects, config)
logger.info('Training model')
cluster.train()
pred = cluster.predictions
logger.info('Done training network')
from muon.project.images import Random_Images
if new:
images = Random_Images.new(cluster)
else:
images = Random_Images.load_group(0)
interact(locals())
def new(config, width, size, permutations, save):
config = Config.load(config)
subjects = pickle.load(open(config.subjects, 'rb'))
cluster = Cluster.create(subjects, config)
logger.info('Training model')
cluster.train()
logger.info('Done training network')
kwargs = {}
if width:
kwargs['width'] = width
if size:
kwargs['image_size'] = size
if permutations:
kwargs['permutations'] = permutations
images = Random_Images.new(cluster, **kwargs)
if save:
images.save_group()
interact(locals())
import csv
logger = logging.getLogger(__name__)
config = 'mnt/dec/dec_no_labels/config_jupyter.json'
config = Config.load(config)
print(config.__dict__)
subjects = pickle.load(open(config.subjects, 'rb'))
# cluster = Cluster.create(subjects, config)
import muon.project.parse_export as pe
agg = pe.Aggregate.load('mh2')
_s = list(agg.data['subjects'].keys())
subjects2 = subjects.subset(_s)
agg.apply_labels(subjects2)
cluster = Cluster.create(subjects2, config)
print('Initializing model')
cluster.initialize()
def load_set(fname):
with open(fname, 'r') as file:
reader = csv.DictReader(file)
return [int(item['subject']) for item in reader]
train = load_set('mnt/training_set/train.csv')
validate = load_set('mnt/training_set/test.csv')
kwargs = {'mode': 'weightedagg', 'batched': False, 'shuffle': False}
mapping = Mapping(cluster, agg, train, validate, **kwargs)