def test_predict(self):
seed = 666
# negative points belong to class 1, positives to 0
p1, p2, p3, p4 = [1, 2], [2, 1], [-1, -2], [-2, -1]
x = ds.array(np.array([p1, p4, p3, p2]), (2, 2))
y = ds.array(np.array([0, 1, 1, 0]).reshape(-1, 1), (2, 1))
csvm = CascadeSVM(cascade_arity=3,
max_iter=10,
tol=1e-4,
kernel='linear',
c=2,
gamma=0.1,
check_convergence=False,
random_state=seed,
verbose=False)
csvm.fit(x, y)
# p5 should belong to class 0, p6 to class 1
p5, p6 = np.array([1, 1]), np.array([-1, -1])
x_test = ds.array(np.array([p1, p2, p3, p4, p5, p6]), (2, 2))
y_pred = csvm.predict(x_test)
l1, l2, l3, l4, l5, l6 = y_pred.collect()
self.assertTrue(l1 == l2 == l5 == 0)
self.assertTrue(l3 == l4 == l6 == 1)
def test_score(self, collect):
seed = 666
# negative points belong to class 1, positives to 0
p1, p2, p3, p4 = [1, 2], [2, 1], [-1, -2], [-2, -1]
x = ds.array(np.array([p1, p4, p3, p2]), (2, 2))
y = ds.array(np.array([0, 1, 1, 0]).reshape(-1, 1), (2, 1))
csvm = CascadeSVM(cascade_arity=3,
max_iter=10,
tol=1e-4,
kernel='rbf',
c=2,
gamma=0.1,
check_convergence=True,
random_state=seed,
verbose=False)
csvm.fit(x, y)
# points are separable, scoring the training dataset should have 100%
# accuracy
x_test = ds.array(np.array([p1, p2, p3, p4]), (2, 2))
y_test = ds.array(np.array([0, 0, 1, 1]).reshape(-1, 1), (2, 1))
accuracy = csvm.score(x_test, y_test, collect)
if not collect:
accuracy = compss_wait_on(accuracy)
self.assertEqual(accuracy, 1.0)
def test_init_params(self):
# Test all parameters with rbf kernel
cascade_arity = 3
max_iter = 1
tol = 1e-4
kernel = 'rbf'
c = 2
gamma = 0.1
check_convergence = True
seed = 666
verbose = False
csvm = CascadeSVM(cascade_arity=cascade_arity,
max_iter=max_iter,
tol=tol,
kernel=kernel,
c=c,
gamma=gamma,
check_convergence=check_convergence,
random_state=seed,
verbose=verbose)
self.assertEqual(csvm._arity, cascade_arity)
self.assertEqual(csvm._max_iter, max_iter)
self.assertEqual(csvm._tol, tol)
self.assertEqual(csvm._clf_params['kernel'], kernel)
self.assertEqual(csvm._clf_params['C'], c)
self.assertEqual(csvm._clf_params['gamma'], gamma)
self.assertEqual(csvm._check_convergence, check_convergence)
self.assertEqual(csvm._verbose, verbose)
# test correct linear kernel and c param (other's are not changed)
kernel, c = 'linear', 0.3
csvm = CascadeSVM(kernel=kernel, c=c)
self.assertEqual(csvm._clf_params['kernel'], kernel)
self.assertEqual(csvm._clf_params['C'], c)
def test_fit(self):
seed = 666
file_ = "tests/files/libsvm/2"
x, y = ds.load_svmlight_file(file_, (10, 300), 780, False)
csvm = CascadeSVM(cascade_arity=3,
max_iter=5,
tol=1e-4,
kernel='linear',
c=2,
gamma=0.1,
check_convergence=True,
random_state=seed,
verbose=False)
csvm.fit(x, y)
self.assertTrue(csvm.converged)
csvm = CascadeSVM(cascade_arity=3,
max_iter=1,
tol=1e-4,
kernel='linear',
c=2,
gamma=0.1,
check_convergence=False,
random_state=seed,
verbose=False)
csvm.fit(x, y)
self.assertFalse(csvm.converged)
self.assertEqual(csvm.iterations, 1)
def test_fit_default_gamma(self):
""" Tests that the fit method converges when using gamma=auto on a
toy dataset """
seed = 666
file_ = "tests/files/libsvm/2"
x, y = ds.load_svmlight_file(file_, (10, 300), 780, False)
csvm = CascadeSVM(cascade_arity=3,
max_iter=5,
tol=1e-4,
kernel='linear',
c=2,
check_convergence=True,
random_state=seed,
verbose=False)
csvm.fit(x, y)
self.assertTrue(csvm.converged)
csvm = CascadeSVM(cascade_arity=3,
max_iter=1,
tol=1e-4,
kernel='linear',
c=2,
gamma=0.1,
check_convergence=False,
random_state=seed,
verbose=False)
csvm.fit(x, y)
self.assertFalse(csvm.converged)
self.assertEqual(csvm.iterations, 1)
def test_init_params(self):
""" Test constructor parameters"""
cascade_arity = 3
max_iter = 1
tol = 1e-4
kernel = 'rbf'
c = 2
gamma = 0.1
check_convergence = True
seed = 666
verbose = False
csvm = CascadeSVM(cascade_arity=cascade_arity,
max_iter=max_iter,
tol=tol,
kernel=kernel,
c=c,
gamma=gamma,
check_convergence=check_convergence,
random_state=seed,
verbose=verbose)
self.assertEqual(csvm.cascade_arity, cascade_arity)
self.assertEqual(csvm.max_iter, max_iter)
self.assertEqual(csvm.tol, tol)
self.assertEqual(csvm.kernel, kernel)
self.assertEqual(csvm.c, c)
self.assertEqual(csvm.gamma, gamma)
self.assertEqual(csvm.check_convergence, check_convergence)
self.assertEqual(csvm.random_state, seed)
self.assertEqual(csvm.verbose, verbose)
def test_fit(self):
"""Tests RandomizedSearchCV fit()."""
x_np, y_np = datasets.load_iris(return_X_y=True)
p = np.random.permutation(len(x_np)) # Pre-shuffling required for CSVM
x = ds.array(x_np[p], (30, 4))
y = ds.array((y_np[p] == 0)[:, np.newaxis], (30, 1))
param_distributions = {'c': stats.expon(scale=0.5),
'gamma': stats.expon(scale=1)}
csvm = CascadeSVM()
n_iter = 12
k = 3
searcher = RandomizedSearchCV(estimator=csvm,
param_distributions=param_distributions,
n_iter=n_iter, cv=k, random_state=0)
searcher.fit(x, y)
expected_keys = {'param_c', 'param_gamma', 'params', 'mean_test_score',
'std_test_score', 'rank_test_score'}
split_keys = {'split%d_test_score' % i for i in range(k)}
expected_keys.update(split_keys)
self.assertSetEqual(set(searcher.cv_results_.keys()), expected_keys)
self.assertEqual(len(searcher.cv_results_['param_c']), n_iter)
self.assertTrue(hasattr(searcher, 'best_estimator_'))
self.assertTrue(hasattr(searcher, 'best_score_'))
self.assertTrue(hasattr(searcher, 'best_params_'))
self.assertTrue(hasattr(searcher, 'best_index_'))
self.assertTrue(hasattr(searcher, 'scorer_'))
self.assertEqual(searcher.n_splits_, k)
def main():
x_ij, y_ij = ds.load_svmlight_file(
"/fefs/scratch/bsc19/bsc19029/PERFORMANCE/datasets/train",
block_size=(5000, 22), n_features=22, store_sparse=True)
csvm = CascadeSVM(c=10000, gamma=0.01)
performance.measure("CSVM", "ijcnn1", csvm.fit, x_ij, y_ij)
def main():
x_kdd = ds.load_txt_file(
"/fefs/scratch/bsc19/bsc19029/PERFORMANCE/datasets/train.csv",
block_size=(11482, 122))
x_kdd = shuffle(x_kdd)
y_kdd = x_kdd[:, 121:122]
x_kdd = x_kdd[:, :121]
csvm = CascadeSVM(c=10000, gamma=0.01)
performance.measure("CSVM", "KDD99", csvm.fit, x_kdd, y_kdd)
def main():
x_kdd = ds.load_txt_file(
"/gpfs/projects/bsc19/COMPSs_DATASETS/dislib/kdd99/train.csv",
block_size=(11482, 122))
x_kdd = shuffle(x_kdd)
y_kdd = x_kdd[:, 121:122]
x_kdd = x_kdd[:, :121]
x_ij, y_ij = ds.load_svmlight_file(
"/gpfs/projects/bsc19/COMPSs_DATASETS/dislib/ijcnn1/train",
block_size=(5000, 22), n_features=22, store_sparse=True)
csvm = CascadeSVM(c=10000, gamma=0.01)
performance.measure("CSVM", "KDD99", csvm.fit, x_kdd, y_kdd)
performance.measure("CSVM", "ijcnn1", csvm.fit, x_ij, y_ij)
def test_fit_2(self):
"""Tests GridSearchCV fit() with different data."""
x_np, y_np = datasets.load_breast_cancer(return_X_y=True)
x = ds.array(x_np, block_size=(100, 10))
x = StandardScaler().fit_transform(x)
y = ds.array(y_np.reshape(-1, 1), block_size=(100, 1))
parameters = {'c': [0.1], 'gamma': [0.1]}
csvm = CascadeSVM()
searcher = GridSearchCV(csvm, parameters, cv=5)
searcher.fit(x, y)
self.assertTrue(hasattr(searcher, 'best_estimator_'))
self.assertTrue(hasattr(searcher, 'best_score_'))
self.assertTrue(hasattr(searcher, 'best_params_'))
self.assertTrue(hasattr(searcher, 'best_index_'))
self.assertTrue(hasattr(searcher, 'scorer_'))
self.assertEqual(searcher.n_splits_, 5)
def test_fit_private_params(self):
kernel = 'rbf'
c = 2
gamma = 0.1
seed = 666
file_ = "tests/files/libsvm/2"
x, y = ds.load_svmlight_file(file_, (10, 300), 780, False)
csvm = CascadeSVM(kernel=kernel, c=c, gamma=gamma, random_state=seed)
csvm.fit(x, y)
self.assertEqual(csvm._clf_params['kernel'], kernel)
self.assertEqual(csvm._clf_params['C'], c)
self.assertEqual(csvm._clf_params['gamma'], gamma)
kernel, c = 'linear', 0.3
csvm = CascadeSVM(kernel=kernel, c=c, random_state=seed)
csvm.fit(x, y)
self.assertEqual(csvm._clf_params['kernel'], kernel)
self.assertEqual(csvm._clf_params['C'], c)
def test_duplicates(self):
""" Tests that C-SVM does not generate duplicate support vectors """
x = ds.array(
np.array([[0, 1], [1, 1], [0, 1], [1, 2], [0, 0], [2, 2], [2, 1],
[1, 0]]), (2, 2))
y = ds.array(np.array([1, 0, 1, 0, 1, 0, 0, 1]).reshape(-1, 1), (2, 1))
csvm = CascadeSVM(c=1, random_state=1, max_iter=100, tol=0)
csvm.fit(x, y)
csvm._collect_clf()
self.assertEqual(csvm._clf.support_vectors_.shape[0], 6)
def test_refit_false(self):
"""Tests GridSearchCV fit() with refit=False."""
x_np, y_np = datasets.load_iris(return_X_y=True)
x = ds.array(x_np, (30, 4))
y = ds.array(y_np[:, np.newaxis], (30, 1))
seed = 0
x, y = shuffle(x, y, random_state=seed)
param_grid = {'max_iter': range(1, 5)}
csvm = CascadeSVM(check_convergence=False)
searcher = GridSearchCV(csvm, param_grid, cv=3, refit=False)
searcher.fit(x, y)
self.assertFalse(hasattr(searcher, 'best_estimator_'))
self.assertTrue(hasattr(searcher, 'best_score_'))
self.assertTrue(hasattr(searcher, 'best_params_'))
self.assertTrue(hasattr(searcher, 'best_index_'))
self.assertTrue(hasattr(searcher, 'scorer_'))
self.assertEqual(searcher.n_splits_, 3)
def test_sparse(self):
""" Tests that C-SVM produces the same results with sparse and dense
data"""
seed = 666
train = "tests/files/libsvm/3"
x_sp, y_sp = ds.load_svmlight_file(train, (10, 300), 780, True)
x_d, y_d = ds.load_svmlight_file(train, (10, 300), 780, False)
csvm_sp = CascadeSVM(random_state=seed)
csvm_sp.fit(x_sp, y_sp)
csvm_d = CascadeSVM(random_state=seed)
csvm_d.fit(x_d, y_d)
sv_d = csvm_d._clf.support_vectors_
sv_sp = csvm_sp._clf.support_vectors_.toarray()
self.assertTrue(np.array_equal(sv_d, sv_sp))
coef_d = csvm_d._clf.dual_coef_
coef_sp = csvm_sp._clf.dual_coef_.toarray()
self.assertTrue(np.array_equal(coef_d, coef_sp))
def test_decision_func(self):
seed = 666
# negative points belong to class 1, positives to 0
# all points are in the x-axis
p1, p2, p3, p4 = [0, 2], [0, 1], [0, -2], [0, -1]
x = ds.array(np.array([p1, p4, p3, p2]), (2, 2))
y = ds.array(np.array([0, 1, 1, 0]).reshape(-1, 1), (2, 1))
csvm = CascadeSVM(cascade_arity=3,
max_iter=10,
tol=1e-4,
kernel='rbf',
c=2,
gamma=0.1,
check_convergence=False,
random_state=seed,
verbose=False)
csvm.fit(x, y)
# p1 should be equidistant to p3, and p2 to p4
x_test = ds.array(np.array([p1, p2, p3, p4]), (2, 2))
y_pred = csvm.decision_function(x_test)
d1, d2, d3, d4 = y_pred.collect()
self.assertTrue(np.isclose(abs(d1) - abs(d3), 0))
self.assertTrue(np.isclose(abs(d2) - abs(d4), 0))
# p5 and p6 should be in the decision function (distance=0)
p5, p6 = np.array([1, 0]), np.array([-1, 0])
x_test = ds.array(np.array([p5, p6]), (1, 2))
y_pred = csvm.decision_function(x_test)
d5, d6 = y_pred.collect()
self.assertTrue(np.isclose(d5, 0))
self.assertTrue(np.isclose(d6, 0))
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--svmlight", help="read files in SVMLight format",
action="store_true")
parser.add_argument("-dt", "--detailed_times",
help="get detailed execution times (read and fit)",
action="store_true")
parser.add_argument("-k", "--kernel", metavar="KERNEL", type=str,
help="linear or rbf (default is rbf)",
choices=["linear", "rbf"], default="rbf")
parser.add_argument("-a", "--arity", metavar="CASCADE_ARITY", type=int,
help="default is 2", default=2)
parser.add_argument("-b", "--block_size", metavar="BLOCK_SIZE", type=str,
help="two comma separated ints that represent the "
"size of the blocks in which to divide the input "
"data (default is 100,100)",
default="100,100")
parser.add_argument("-i", "--iteration", metavar="MAX_ITERATIONS",
type=int, help="default is 5", default=5)
parser.add_argument("-g", "--gamma", metavar="GAMMA", type=float,
help="(only for rbf kernel) default is 1 / n_features",
default=None)
parser.add_argument("-c", metavar="C", type=float, default=1,
help="Penalty parameter C of the error term. "
"Default:1")
parser.add_argument("-f", "--features", metavar="N_FEATURES",
help="number of features of the input data "
"(only for SVMLight files)",
type=int, default=None, required=False)
parser.add_argument("-t", "--test-file", metavar="TEST_FILE_PATH",
help="test file path", type=str, required=False)
parser.add_argument("-o", "--output_file", metavar="OUTPUT_FILE_PATH",
help="output file path", type=str, required=False)
parser.add_argument("--convergence", help="check for convergence",
action="store_true")
parser.add_argument("--dense", help="store data in dense format (only "
"for SVMLight files)",
action="store_true")
parser.add_argument("train_data",
help="input file in CSV or SVMLight format", type=str)
parser.add_argument("-v", "--verbose", action="store_true")
parser.add_argument("-s", "--shuffle", help="shuffle input data",
action="store_true")
args = parser.parse_args()
train_data = args.train_data
s_time = time.time()
read_time = 0
if not args.gamma:
gamma = "auto"
else:
gamma = args.gamma
sparse = not args.dense
bsize = args.block_size.split(",")
block_size = (int(bsize[0]), int(bsize[1]))
if args.svmlight:
x, y = ds.load_svmlight_file(train_data, block_size, args.features,
sparse)
else:
x = ds.load_txt_file(train_data, block_size)
y = x[:, x.shape[1] - 2: x.shape[1] - 1]
x = x[:, :x.shape[1] - 1]
if args.shuffle:
x, y = shuffle(x, y)
if args.detailed_times:
barrier()
read_time = time.time() - s_time
s_time = time.time()
csvm = CascadeSVM(cascade_arity=args.arity, max_iter=args.iteration,
c=args.c, gamma=gamma,
check_convergence=args.convergence, verbose=args.verbose)
csvm.fit(x, y)
barrier()
fit_time = time.time() - s_time
out = [args.kernel, args.arity, args.part_size, csvm._clf_params["gamma"],
args.c, csvm.iterations, csvm.converged, read_time, fit_time]
if os.path.isdir(train_data):
n_files = os.listdir(train_data)
out.append(len(n_files))
if args.test_file:
if args.svmlight:
x_test, y_test = ds.load_svmlight_file(args.test_file, block_size,
args.features,
sparse)
else:
x_test = ds.load_txt_file(args.test_file, block_size)
y_test = x_test[:, x_test.shape[1] - 1: x_test.shape[1]]
x_test = x_test[:, :x_test.shape[1] - 1]
out.append(compss_wait_on(csvm.score(x_test, y_test)))
if args.output_file:
with open(args.output_file, "ab") as f:
wr = csv.writer(f)
wr.writerow(out)
else:
print(out)
def main():
h = .02 # step size in the mesh
names = ["Linear C-SVM", "RBF C-SVM", "Random forest"]
classifiers = [
CascadeSVM(kernel="linear", c=0.025, max_iter=5),
CascadeSVM(gamma=2, c=1, max_iter=5),
RandomForestClassifier(random_state=1)
]
x, y = make_classification(n_features=2,
n_redundant=0,
n_informative=2,
random_state=1,
n_clusters_per_class=1)
rng = np.random.RandomState(2)
x += 2 * rng.uniform(size=x.shape)
linearly_separable = (x, y)
datasets = [
make_moons(noise=0.3, random_state=0),
make_circles(noise=0.2, factor=0.5, random_state=1), linearly_separable
]
preprocessed_data = dict()
scores = dict()
mesh_accuracy_ds = dict()
for ds_cnt, data in enumerate(datasets):
# preprocess dataset, split into training and test part
x, y = data
x = StandardScaler().fit_transform(x)
x_train, x_test, y_train, y_test = \
train_test_split(x, y, test_size=.4, random_state=42)
ds_x_train = ds.array(x_train, block_size=(20, 2))
ds_y_train = ds.array(y_train.reshape(-1, 1), block_size=(20, 1))
ds_x_test = ds.array(x_test, block_size=(20, 2))
ds_y_test = ds.array(y_test.reshape(-1, 1), block_size=(20, 1))
x_min, x_max = x[:, 0].min() - .5, x[:, 0].max() + .5
y_min, y_max = x[:, 1].min() - .5, x[:, 1].max() + .5
xx, yy = np.meshgrid(np.arange(x_min, x_max, h),
np.arange(y_min, y_max, h))
preprocessed_data[ds_cnt] = x, x_train, x_test, y_train, y_test, xx, yy
for name, clf in zip(names, classifiers):
clf.fit(ds_x_train, ds_y_train)
scores[(ds_cnt, name)] = clf.score(ds_x_test, ds_y_test)
mesh = np.c_[xx.ravel(), yy.ravel()]
mesh_array = ds.array(mesh, (mesh.shape[0], 2))
if hasattr(clf, "decision_function"):
mesh_proba = clf.decision_function(mesh_array)
else:
mesh_proba = clf.predict_proba(mesh_array)
mesh_accuracy_ds[(ds_cnt, name)] = mesh_proba
# Synchronize while plotting the results
plt.figure(figsize=(27, 9))
i = 1
for ds_cnt, data in enumerate(datasets):
x, x_train, x_test, y_train, y_test, xx, yy = preprocessed_data[ds_cnt]
# just plot the dataset first
cm = plt.cm.RdBu
cm_bright = ListedColormap(['#FF0000', '#0000FF'])
ax = plt.subplot(len(datasets), len(classifiers) + 1, i)
if ds_cnt == 0:
ax.set_title("Input data")
# Plot the training points
ax.scatter(x_train[:, 0],
x_train[:, 1],
c=y_train,
cmap=cm_bright,
edgecolors='k')
# Plot the testing points
ax.scatter(x_test[:, 0],
x_test[:, 1],
c=y_test,
cmap=cm_bright,
alpha=0.6,
edgecolors='k')
ax.set_xlim(xx.min(), xx.max())
ax.set_ylim(yy.min(), yy.max())
ax.set_xticks(())
ax.set_yticks(())
i += 1
# iterate over classifiers
for name, clf in zip(names, classifiers):
ax = plt.subplot(len(datasets), len(classifiers) + 1, i)
score = compss_wait_on(scores[(ds_cnt, name)])
mesh_proba = mesh_accuracy_ds[(ds_cnt, name)]
if hasattr(clf, "decision_function"):
Z = mesh_proba.collect()
else:
Z = mesh_proba.collect()[:, 1]
# Put the result into a color plot
Z = Z.reshape(xx.shape)
ax.contourf(xx, yy, Z, cmap=cm, alpha=.8)
# Plot the training points
ax.scatter(x_train[:, 0],
x_train[:, 1],
c=y_train,
cmap=cm_bright,
edgecolors='k')
# Plot the testing points
ax.scatter(x_test[:, 0],
x_test[:, 1],
c=y_test,
cmap=cm_bright,
edgecolors='k',
alpha=0.6)
ax.set_xlim(xx.min(), xx.max())
ax.set_ylim(yy.min(), yy.max())
ax.set_xticks(())
ax.set_yticks(())
if ds_cnt == 0:
ax.set_title(name)
ax.text(xx.max() - .3,
yy.min() + .3, ('%.2f' % score).lstrip('0'),
size=15,
horizontalalignment='right')
i += 1
plt.tight_layout()
plt.show()