def main():
lambdas = [10**i for i in range(-6,2)] + [0]
full_data = load_full_data()
images_all = full_data['images']
labels_all = full_data['labels']
thetas = {}
trained = []
for i in range(10):
for j in range(i+1,10):
now = datetime.datetime.now().strftime('%c')
#print(f"[{datetime.datetime.now().strftime('%c')}] digits {i} vs. {j}")
print("[%s] digits %d vs %d"%(now, i, j))
filter = ((labels_all == i) | (labels_all == j))[:, 0]
images = images_all[filter]
labels = ((labels_all[filter]) == i)
n,d = images.shape
X = np.concatenate((images, np.ones((images.shape[0],1))),axis=1)
y = labels
phi = np.random.uniform(size=(d+1, 1))
thetas = [{'first_digit':i,
'second_digit':j,
'theta': m.learn_stochastic(X, y, l),
'lambda': l}
for l in lambdas]
trained.append(thetas)
save_data("trained.npz", trained=trained)
save_data("trained.npz", trained=trained)
def main():
lambdas = [10**i for i in range(-6, 2)] + [0]
data = shared.load_full_data()
X, y = data['images'], data['labels']
thetas = []
for l in lambdas:
print("Lambda = %.0e" % l)
thetas.append({'theta': moo.soft_run(X, y, l),
'lambda': l})
shared.save_data("trained_multinomial.npz", trained=thetas)
def save_parameters(theta):
save_data('params', theta=best_theta)
def run():
data = load_full_data()
small_data = get_subset_of_data(data)
save_data("small_data", **small_data) #images=small_data['images'], labels=small_data['labels'])
