def compute_train_kernel(self,
g,
m,
t=20,
approx=True,
I=100,
delta=0.025,
skip_variance=False):
kernel = FastSK(g=g,
m=m,
t=t,
approx=approx,
max_iters=I,
delta=delta,
skip_variance=skip_variance)
kernel.compute_train(self.train_seq)
def train_and_test(self,
g,
m,
t,
approx,
I=100,
delta=0.025,
skip_variance=False,
C=1):
kernel = FastSK(
g=g,
m=m,
t=t,
approx=approx,
max_iters=I,
delta=delta,
skip_variance=skip_variance,
)
kernel.compute_kernel(self.train_seq, self.test_seq)
self.Xtrain = kernel.get_train_kernel()
self.Xtest = kernel.get_test_kernel()
self.stdevs = kernel.get_stdevs()
svm = LinearSVC(C=C, class_weight="balanced")
self.clf = CalibratedClassifierCV(svm,
cv=5).fit(self.Xtrain, self.Ytrain)
acc, auc = self.evaluate_clf()
return acc, auc
def main(args):
## Get data
reader = FastaUtility()
Xtrain, Ytrain = reader.read_data(args.train)
Xtest, Ytest = reader.read_data(args.test)
Ytest = np.array(Ytest).reshape(-1, 1)
## Compute kernel matrix
fastsk = FastSK(g=10, m=6, t=1, approx=True)
fastsk.compute_kernel(Xtrain, Xtest)
Xtrain = fastsk.get_train_kernel()
Xtest = fastsk.get_test_kernel()
reader = FastaUtility()
Xseq, Ytrain = reader.read_data(args.train)
## Train a linear SVM
svm = LinearSVC(C=1)
clf = CalibratedClassifierCV(svm, cv=5).fit(Xtrain, Ytrain)
## Evaluate
acc = clf.score(Xtest, Ytest)
probs = clf.predict_proba(Xtest)[:, 1]
auc = roc_auc_score(Ytest, probs)
print("Linear SVM:\n\tAcc = {}, AUC = {}".format(acc, auc))
assert auc >= 0.9, "AUC is not correct. Should be >= 0.9. Received: {}".format(auc)
def train_and_test(self,
g,
m,
t,
approx,
I=100,
delta=0.025,
skip_variance=False):
kernel = FastSK(
g=g,
m=m,
t=t,
approx=approx,
max_iters=I,
delta=delta,
skip_variance=skip_variance,
)
kernel.compute_kernel(self.train_seq, self.test_seq)
self.Xtest = kernel.get_test_kernel()
self.Xtest = np.array(self.Xtest).reshape(len(self.Xtest), -1)
self.Xtrain = kernel.get_train_kernel()
self.Xtrain = np.array(self.Xtrain).reshape(len(self.Xtrain), -1)
# Can replace Lasso with alternative regression approaches such as SVR
model = LassoCV(cv=5, n_jobs=t,
random_state=293).fit(self.Xtrain, self.Ytrain)
r2 = model.score(self.Xtest, self.Ytest)
return r2
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--train',
default='../data/EP300.train.fasta',
help='training sequences file')
parser.add_argument('--test',
default='../data/EP300.test.fasta',
help='test sequences file')
args = parser.parse_args()
## Compute kernel matrix
fastsk = FastSK(g=10, m=6, t=1, approx=True)
fastsk.compute_kernel(args.train, args.test)
Xtrain = fastsk.get_train_kernel()
Xtest = fastsk.get_test_kernel()
reader = FastaUtility()
Xseq, Ytrain = reader.read_data(args.train)
## Use linear SVM
svm = LinearSVC(C=1)
clf = CalibratedClassifierCV(svm, cv=5).fit(Xtrain, Ytrain)
## Evaluate
reader = FastaUtility()
Xseq, Ytest = reader.read_data(args.test)
args.t,
args.approx,
args.I,
args.delta,
)
skip_variance = args.skip_variance
### Read the data
reader = FastaUtility()
Xtrain, Ytrain = reader.read_data(train_file)
Xtest, Ytest = reader.read_data(test_file)
Ytest = np.array(Ytest).reshape(-1, 1)
### Compute the fastsk kernel
start = time.time()
fastsk = FastSK(
g=g, m=m, t=t, approx=approx, max_iters=I, delta=d, skip_variance=skip_variance
)
fastsk.compute_kernel(Xtrain, Xtest)
end = time.time()
print("Kernel computation time: ", end - start)
Xtrain = fastsk.get_train_kernel()
Xtest = fastsk.get_test_kernel()
### Use linear SVM
svm = LinearSVC(C=C)
clf = CalibratedClassifierCV(svm, cv=5).fit(Xtrain, Ytrain)
acc, auc = evaluate_clf(clf, Xtest, Ytest)
print("Linear SVM:\n\tAcc = {}, AUC = {}".format(acc, auc))
import glob
from fastsk import FastSK
datasets = []
for filepath in glob.iglob('../../data/*.train.fasta'):
s = filepath[:filepath.find(".train")]
datasets.append(s)
#print(s)
for s in datasets:
name = s[s.find("/") + 1:]
# Uncomment for confirmation for each dataset train
'''
s = input("about to train on " + name + " dataset.\ny/n?")
if s.lower() != "y":
continue
'''
fastsk = FastSK(g=7, m=3, t=16, approx=True, max_iters=75)
fastsk.compute_kernel(Xtrain=s + ".train.fasta",
Xtest="../../data/" + name + ".test.fasta")
fastsk.fit()
fastsk.score(metric='auc')