def make_methods_list():
methods = []
"""
# SVMs.
for C in [1e2, 1e3, 1e4]:
for gamma in [0.01, 0.05, 0.1]:
methods.append(delayed(sl.svm.SVC)(kernel='rbf', C=C, gamma=gamma))
methods.append(delayed(sl.svm.SVC)(kernel='poly', degree=2, C=C,
gamma=gamma))
methods.append(delayed(sl.svm.SVC)(kernel='poly', degree=3, C=C,
gamma=gamma))
methods.append(delayed(sl.svm.SVC)(kernel='linear', C=C))
# Misc.
methods.append(delayed(sl.tree.DecisionTreeClassifier)())
methods.append(delayed(sl.naive_bayes.GaussianNB)())
"""
# Random Forests.
for n in [10, 50, 100, 200, 500]:
methods.append(delayed(sl.ensemble.RandomForestClassifier)
(n_estimators=n))
for n in [10, 50, 100, 200, 500]:
methods.append(delayed(sl.ensemble.RandomForestClassifier)
(n_estimators=n, criterion='entropy'))
# Gradient boosting.
for n in [10, 50, 100, 200, 500]:
for a in [.0001, .001, .01, .1, 1., 10.]:
methods.append(delayed(sl.ensemble.GradientBoostingClassifier)
(n_estimators=n, learn_rate=a))
# Nearest neighbors.
for n in [1, 5, 10, 15, 20]:
methods.append(delayed(sl.neighbors.KNeighborsClassifier)
(n_neighbors=n))
for n in [1, 5, 10, 15, 20]:
methods.append(delayed(sl.neighbors.KNeighborsClassifier)
(n_neighbors=n, weights='distance'))
# l1-penalized logistic regression
for c in [0.0001, 0.001, 0.01, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100]:
methods.append(delayed(sl.linear_model.LogisticRegression)
(C=c, penalty='l1', tol=0.01))
# l2-penalized logistic regression
for c in [0.0001, 0.001, 0.01, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100]:
methods.append(delayed(sl.linear_model.LogisticRegression)
(C=c, penalty='l2', tol=0.01))
return methods
if args.j < 0 or args.j >= n_jobs:
raise ValueError('Job ID out of range.')
# Range of jobs to run
job_batchsize = int(J / n_jobs)
a = args.j * job_batchsize
b = a + job_batchsize if (args.j < n_jobs-1) else J
# Get cross-validation folds
cv = LeaveOneOut(n) if (args.k == 0) else KFold(n, args.k)
if not args.a:
# Setup list of jobs
jobs = iter(delayed(run_method)(method, X, y, train, test, force=args.f)
for method, (train, test) in itertools.product(methods, cv))
# Run only jobs in batch
for job in itertools.islice(jobs, a, b):
run_delayed(job)
else: # Aggregate results
accs = np.empty(M * args.k)
wall = np.empty(M * args.k)
accs.fill(np.nan)
wall.fill(np.nan)
# Fetch data
jobs = iter(delayed(run_method)(method, X, y, train, test, load=True)
for method, (train, test) in itertools.product(methods, cv))
