def loss(self, model, preprocess=None, task=None):
"""Run Loss Statistics
Parameters
----------
model : class with predict method
Model to test
preprocess : np.array -> np.array
Preprocessing method to run; if None, no preprocessing is performed
task : Task
Task to register the test under
"""
if task is not None:
task = Task()
task.start(name='Tester', desc=self.desc)
# Predict
if preprocess is not None:
predicted = model.predict(
np.array([preprocess(x) for x in self.data]))
else:
predicted = model.predict(self.data)
# Compute error
self.error = self.__error(predicted, self.labels)
# Print stats
for key, value in self.error.items():
task.print("{label}: {val}".format(label=LABELS[key], val=value))
task.done(desc='Done running tests.')
task.print('Time per test: {t}s'.format(t=task.runtime() /
self.error['total']))
def __new(self, task):
"""Create new W and b"""
if task is None:
task = Task()
task.start(name='Random Fourier Feature', desc=self.__str__())
# Create feature
self.create()
task.done(self.W,
self.b,
desc="{desc} created".format(desc=self.__str__()))
def __new(self, task, cores):
if task is None:
task = Task()
task.start(name='Random Binning Feature', desc=self.__str__())
gen = task.pool(get_p_set, [self.d for _ in range(self.D)],
cores=cores,
process=True,
name='Random Binning Feature')
self.delta = np.array([x[0] for x in gen], dtype=np.float32)
self.mu = np.array([x[1] for x in gen], dtype=np.float32)
task.done(self.delta,
self.mu,
desc="{desc} created".format(desc=self.__str__()))
def __init__(self,
patients,
transform=[None, None],
feature=None,
cores=None,
p=1,
task=None,
process=False):
if task is None:
task = Task()
task.start(name='IDC Dataset', desc='Loading Images...')
if transform is None:
transform = [None, None]
if process:
self.data, self.classes = task.pool(
proc_wrapper,
patients,
process=True,
shared_args=[transform[0], transform[1], p, feature],
shared_init=pinit,
name='Loader',
recursive=True,
reducer=proc_reducer,
cores=cores)
else:
self.data, self.classes = task.pool(load_patient,
patients,
process=False,
shared_kwargs={
'transform': transform,
'feature': feature,
'p': p
},
reducer=reducer,
name='Loader',
recursive=False,
threads=cores)
task.done(self.data,
self.classes,
desc="{n} images ({p}%) sampled from {k} patients".format(
n=self.classes.shape[0], k=len(patients), p=p * 100))
def __init__(self, n, dataset, task=None):
if task is None:
task = Task()
task.start(name='CKM', desc="Color K Means Clustering")
self.n = n
knndata = []
for d in dataset.data:
d = d.reshape(50, 50, 3)
for i in range(10):
x = random.randint(0, 49)
y = random.randint(0, 49)
knndata.append(d[x, y])
self.kmeans = KMeans(n_clusters=n).fit(knndata).cluster_centers_
task.done(desc="Generated K Means model (n={n})".format(n=self.n))
def load_patient(patient, p=1, transform=None, feature=None, task=None):
"""Load a patient
Parameters
----------
args : [str, Task, array]
[0] patient to load
[1] task to register under
[2] [float, class, mixed type[], function]
[0] proportion of samples to load
[1] feature generator class
[2] feature arguments
[3] image transform
"""
# Set up task
if task is None:
task = Task()
task.start(name='Loader', desc='Loading Patient {p}'.format(p=patient))
# Config: pass in p, feature, transform
load_args = {'p': p, 'feature': feature, 'transform': transform}
try:
class_0 = load_images(os.path.join(BASE_PATH, patient, '0'),
**load_args)
class_1 = load_images(os.path.join(BASE_PATH, patient, '1'),
**load_args)
classes = np.concatenate([
np.zeros(class_0.shape[0], dtype=np.int8),
np.ones(class_1.shape[0], dtype=np.int8)
])
data = np.concatenate([class_0, class_1])
task.done(data, classes, desc="loaded patient {p}".format(p=patient))
return (data, classes)
except Exception as e:
task.error("error loading patient {p}: {e}".format(p=patient, e=e))
task.done(desc="could not load patient {p}".format(p=patient))
return (None, None)