batch_mem = tovar(np.array(batch_mem).transpose(0,2,1).reshape(BS,1,FEATURES+CLASSES,NTRAIN)) batch_test = tovar(np.array(batch_test).transpose(0,2,1).reshape(BS,1,FEATURES,100)) batch_label = tovar(np.array(batch_label).transpose(0,2,1)) class_count = torch.cuda.FloatTensor(np.array(class_count)) net.zero_grad() p = net.forward(batch_mem, batch_test, class_count) loss = -torch.sum(p*batch_label,1).mean() loss.backward() net.adam.step() err = loss.cpu().data.numpy()[0] return err net = ClassifierGenerator(FEATURES=2, CLASSES=4, NETSIZE=384).cuda() difficulty_level = 1.0 errs = [] err = 0 err_count = 0 for i in range(40000): err += trainingStep(net, 400, min_difficulty = difficulty_level, max_difficulty = difficulty_level, feature_variation = False, class_variation = False) err_count += 1 if err_count >= 50: err = err/err_count errs.append(err)
plt.imshow(im, extent=[-3, 3, 3, -3])
for j in range(4):
plt.scatter(x[0, 0, 0, yl == j],
x[0, 0, 1, yl == j],
c=colors[j],
edgecolors='k',
lw=1,
s=10)
plt.xticks([])
plt.yticks([])
plt.xlim(-3, 3)
plt.ylim(-3, 3)
net2_4_400_1 = ClassifierGenerator(2, 4, 384).cuda()
net2_4_400_1.load_state_dict(
torch.load("models/classifier-generator-2-4-base.pth"))
net2_4_20_1 = ClassifierGenerator(2, 4, 384).cuda()
net2_4_20_1.load_state_dict(
torch.load("models/classifier-generator-2-4-N20.pth"))
net2_4_100_1 = ClassifierGenerator(2, 4, 384).cuda()
net2_4_100_1.load_state_dict(
torch.load("models/classifier-generator-2-4-N100.pth"))
net2_4_100_4 = ClassifierGenerator(2, 4, 384).cuda()
net2_4_100_4.load_state_dict(
torch.load("models/classifier-generator-2-4-diff4.pth"))
return err
data_names = []
data_x = []
data_y = []
for file in glob.glob("data/*.npz"):
data = np.load(file)
if np.unique(data['y']).shape[0] <= 16:
data_names.append(file[5:-4])
data_x.append(data['x'].copy())
data_y.append(data['y'].copy().astype(np.int32))
for didx in range(len(data_names)):
net = ClassifierGenerator(FEATURES=128, CLASSES=16, NETSIZE=384).cuda()
net.load_state_dict(torch.load("models/classifier-generator-128-16.pth"))
tdx = []
tdy = []
for didx2 in range(len(data_names)):
if didx2 != didx:
if data_x[didx2].shape[0] >= 120:
tdx.append(data_x[didx2])
tdy.append(data_y[didx2])
for i in range(20):
err = trainingStep(net, 100, 20, tdx, tdy)
f = open("training_curves/finetuning-%s.txt" % data_names[didx], "a")
f.write("%d %.6g\n" % (i, err))
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import roc_auc_score
import xgboost as xgb
import warnings
def fxn():
warnings.warn("deprecated", DeprecationWarning)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
fxn()
net128_16 = ClassifierGenerator(128, 16, NETSIZE=384).cuda()
net128_16.load_state_dict(torch.load("models/classifier-generator-128-16.pth"))
net32_16 = ClassifierGenerator(32, 16, NETSIZE=384).cuda()
net32_16.load_state_dict(torch.load("models/classifier-generator-32-16.pth"))
dataset_descriptions = {
"data/immunotherapy.npz": "Immunotherapy\\cite{khozeimeh2017expert, khozeimeh2017intralesional}",
"data/foresttype.npz": "Forest type\\cite{johnson2012using}",
"data/winetype.npz" : "Wine type\\cite{forina1990parvus}",
"data/cryotherapy.npz" : "Cryotherapy\\cite{khozeimeh2017expert, khozeimeh2017intralesional}",
"data/chronic-kidney.npz" : "Chronic kidney\\cite{chronickidney}",
"data/echocardiogram.npz" : "Echocardiogram\\cite{echocardiogram}",
"data/haberman.npz" : "Haberman\\cite{haberman1976generalized}",
"data/iris.npz" : "Iris\\cite{fisher1936use}",
"data/hcc-survival.npz" : "HCC Survival\\cite{santos2015new}",
"data/horse-colic.npz" : "Horse Colic\\cite{horsecolic}",
net.zero_grad()
p = net.forward(batch_mem, batch_test, class_count)
loss = -torch.sum(p * batch_label, 1).mean()
loss.backward()
net.adam.step()
err = loss.cpu().data.numpy()[0]
return err
# Echocardiogram, blood transfusion, autism
echocardio = np.load("data/echocardiogram.npz")
bloodtransfusion = np.load("data/bloodtransfusion.npz")
autism = np.load("data/autism.npz")
net = ClassifierGenerator(FEATURES=128, CLASSES=16, NETSIZE=384).cuda()
difficulty_level = 0.0125
errs = []
err = 0
err_count = 0
for i in range(100000):
err += trainingStep(net,
100,
min_difficulty=difficulty_level * 0.5,
max_difficulty=difficulty_level * 1.5)
err_count += 1
if i % 10000 == 5000:
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import roc_auc_score
import xgboost as xgb
import warnings
def fxn():
warnings.warn("deprecated", DeprecationWarning)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
fxn()
net32_16 = ClassifierGenerator(32, 16, 384).cuda()
net32_16.load_state_dict(torch.load("models/classifier-generator-32-16.pth"))
net128_16 = ClassifierGenerator(128, 16, 384).cuda()
net128_16.load_state_dict(torch.load("models/classifier-generator-128-16.pth"))
methods = [
lambda: SVC(kernel='linear', C=1, probability=True),
lambda: SVC(kernel='rbf', C=1, probability=True), RandomForestClassifier,
lambda: xgb.XGBClassifier(n_jobs=64), KNeighborsClassifier,
lambda: NetworkSKL(net32_16, ensemble=30),
lambda: NetworkSKL(net128_16, ensemble=30)
]
# Sweep Feat
f = open("results/Feat-128-16-100.txt", "wb")
alltrain_x.append(train_x)
alltrain_y.append(train_y)
alltest_x.append(test_x)
alltest_y.append(test_y)
t0 = time.time()
results = [
evalClassifier(m, alltrain_x, alltrain_y, alltest_x, alltest_y)
for m in methods
]
t1 = time.time()
return t1 - t0
net = ClassifierGenerator(128, 16, 384).cpu()
net.load_state_dict(torch.load("models/classifier-generator-128-16.pth"))
netskl = NetworkSKL(net)
methods = [lambda: SVC(kernel='linear', probability=True), \
lambda: SVC(kernel='rbf', probability=True), \
RandomForestClassifier, \
lambda: xgb.XGBClassifier(n_jobs = 12), \
KNeighborsClassifier, \
lambda: NetworkSKL(net,cuda=False), \
lambda: NetworkSKL(net,cuda=True)]
values = [[100, 400], [200, 400], [400, 400], [100, 800], [100, 1600],
[100, 3200], [100, 6400]]
f = open("results/timings.tex", "wb")
batch_test = tovar(
np.array(batch_test).transpose(0, 2, 1).reshape(BS, 1, FEATURES, 100))
batch_label = tovar(np.array(batch_label).transpose(0, 2, 1))
class_count = torch.cuda.FloatTensor(np.array(class_count))
net.zero_grad()
p = net.forward(batch_mem, batch_test, class_count)
loss = -torch.sum(p * batch_label, 1).mean()
loss.backward()
net.adam.step()
err = loss.cpu().data.numpy()[0]
return err
net = ClassifierGenerator(FEATURES=2, CLASSES=4, NETSIZE=384).cuda()
difficulty_level = 1.0
errs = []
err = 0
err_count = 0
for i in range(40000):
err += trainingStep(net,
100,
min_difficulty=difficulty_level,
max_difficulty=difficulty_level,
feature_variation=False,
class_variation=False)
err_count += 1