def fit_lts(X_train, y_train, X_test, y_test, shap_dict, reg, max_it, shap_out_path, pred_out_path, timing_out_path):
# Fit LTS model, print metrics on test-set, write away predictions and shapelets
clf = ShapeletModel(n_shapelets_per_size=shap_dict,
max_iter=max_it, verbose_level=0, batch_size=1,
optimizer='sgd', weight_regularizer=reg)
start = time.time()
clf.fit(
np.reshape(
X_train,
(X_train.shape[0], X_train.shape[1], 1)
),
y_train
)
learning_time = time.time() - start
with open(shap_out_path, 'w+') as ofp:
for shap in clf.shapelets_:
ofp.write(str(np.reshape(shap, (-1))) + '\n')
with open(timing_out_path, 'w+') as ofp:
ofp.write(str(learning_time))
X_distances_train = clf.transform(X_train)
X_distances_test = clf.transform(X_test)
fit_lr(X_distances_train, y_train, X_distances_test, y_test, pred_out_path)
def test_shapelet_lengths():
pytest.importorskip('tensorflow')
from tslearn.shapelets import ShapeletModel
# Test variable-length
y = [0, 1]
time_series = to_time_series_dataset([[1, 2, 3, 4, 5], [3, 2, 1]])
clf = ShapeletModel(n_shapelets_per_size={3: 1},
max_iter=1,
verbose=0,
random_state=0)
clf.fit(time_series, y)
weights_shapelet = [np.array([[1, 2, 3]])]
clf.set_weights(weights_shapelet, layer_name="shapelets_0_0")
tr = clf.transform(time_series)
np.testing.assert_allclose(tr,
np.array([[0.], [8. / 3]]))
# Test max_size to predict longer series than those passed at fit time
y = [0, 1]
time_series = to_time_series_dataset([[1, 2, 3, 4, 5], [3, 2, 1]])
clf = ShapeletModel(n_shapelets_per_size={3: 1},
max_iter=1,
verbose=0,
max_size=6,
random_state=0)
clf.fit(time_series[:, :-1], y) # Fit with size 4
weights_shapelet = [np.array([[1, 2, 3]])]
clf.set_weights(weights_shapelet, layer_name="shapelets_0_0")
tr = clf.transform(time_series)
np.testing.assert_allclose(tr,
np.array([[0.], [8. / 3]]))
def lts_discovery(X_train, y_train, X_test, y_test, nr_shap, l, r, reg, max_it, shap_out_path, pred_out_path, timing_out_path):
# Fit LTS model, print metrics on test-set, write away predictions and shapelets
shapelet_dict = grabocka_params_to_shapelet_size_dict(
X_train.shape[0], X_train.shape[1], int(nr_shap*X_train.shape[1]), l, r
)
clf = ShapeletModel(n_shapelets_per_size=shapelet_dict,
max_iter=max_it, verbose_level=0, batch_size=1,
optimizer='sgd', weight_regularizer=reg)
start = time.time()
clf.fit(
np.reshape(
X_train,
(X_train.shape[0], X_train.shape[1], 1)
),
y_train
)
learning_time = time.time() - start
print('Learning shapelets took {}s'.format(learning_time))
with open(shap_out_path, 'w+') as ofp:
for shap in clf.shapelets_:
ofp.write(str(np.reshape(shap, (-1))) + '\n')
with open(timing_out_path, 'w+') as ofp:
ofp.write(str(learning_time))
X_distances_train = clf.transform(X_train)
X_distances_test = clf.transform(X_test)
fit_lr(X_distances_train, y_train, X_distances_test, y_test, pred_out_path)
def fit(self, X, y):
"""Fit the model using X as training data and y as target values
Parameters
----------
X : {array-like}
Training data. Shape [n_samples, n_features].
y : {array-like, sparse matrix}
Target values of shape = [n_samples] or [n_samples, n_outputs]
"""
self.X = X
self.y = y
n_shapelets_per_size = self.shapelet_model_params.get(
"n_shapelets_per_size", "heuristic")
if n_shapelets_per_size == "heuristic":
n_ts, ts_sz = X.shape[:2]
n_classes = len(set(y))
n_shapelets_per_size = grabocka_params_to_shapelet_size_dict(
n_ts=n_ts,
ts_sz=ts_sz,
n_classes=n_classes,
l=self.shapelet_model_params.get("l", 0.1),
r=self.shapelet_model_params.get("r", 2))
shp_clf = ShapeletModel(
n_shapelets_per_size=n_shapelets_per_size,
optimizer=self.shapelet_model_params.get("optimizer", "sgd"),
weight_regularizer=self.shapelet_model_params.get(
"weight_regularizer", .01),
max_iter=self.shapelet_model_params.get("max_iter", 100),
random_state=self.random_state,
verbose=self.shapelet_model_params.get("verbose", 0))
shp_clf.fit(X, y)
X_transformed = shp_clf.transform(X)
self.X_transformed = X_transformed
if self.tau is not None:
self.X_thresholded = 1 * (self.X_transformed < self.tau)
clf = DecisionTreeClassifier()
param_grid = self.decision_tree_grid_search_params
grid = GridSearchCV(clf,
param_grid=param_grid,
scoring='accuracy',
n_jobs=-1,
verbose=0)
grid.fit(self.X_thresholded, y)
else:
grids = []
grids_scores = []
for quantile in self.tau_quantiles:
_X_thresholded = 1 * (self.X_transformed < (np.quantile(
self.X_transformed, quantile)))
clf = DecisionTreeClassifier()
param_grid = self.decision_tree_grid_search_params
grid = GridSearchCV(clf,
param_grid=param_grid,
scoring='accuracy',
n_jobs=-1,
verbose=0)
grid.fit(_X_thresholded, y)
grids.append(grid)
grids_scores.append(grid.best_score_)
grid = grids[np.argmax(np.array(grids_scores))]
best_quantile = self.tau_quantiles[np.argmax(
np.array(grids_scores))]
self.tau = np.quantile(self.X_transformed, best_quantile)
self.X_thresholded = 1 * (self.X_transformed < self.tau)
clf = DecisionTreeClassifier(**grid.best_params_)
clf.fit(self.X_thresholded, y)
if self.prune_duplicate_tree_leaves:
prune_duplicate_leaves(
clf) # FIXME: does it influence the .tree properties?
self.decision_tree = clf
self.decision_tree_explorable = NewTree(clf)
self.decision_tree_explorable.build_tree()
self._shapelet_model = shp_clf
self._build_tree_graph()
return self
n_iterations = np.random.choice([2000, 5000, 10000])
shapelet_dict = grabocka_params_to_shapelet_size_dict(
X_train.shape[0], X_train.shape[1], int(K * X_train.shape[1]), L,
R)
clf = ShapeletModel(n_shapelets_per_size=shapelet_dict,
max_iter=n_iterations,
verbose_level=0,
batch_size=1,
optimizer='sgd',
weight_regularizer=_lambda)
clf.fit(np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1)),
y_train)
X_distances_train = clf.transform(X_train)
X_distances_test = clf.transform(X_test)
lr = GridSearchCV(LogisticRegression(), {
'penalty': ['l1', 'l2'],
'C': [0.001, 0.01, 0.1, 1.0, 10.0]
})
lr.fit(X_distances_train, y_train)
acc = accuracy_score(y_test, lr.predict(X_distances_test))
print([K, L, R, _lambda, n_iterations], acc)
lts_results.append([K, L, R, _lambda, n_iterations, acc])
# Sample random hyper-parameters for GENDIS
l=0.125,
r=1)
# Define the model and fit it using the training data
shp_clf = ShapeletModel(n_shapelets_per_size=shapelet_sizes,
weight_regularizer=.01,
max_iter=100,
verbose=0,
random_state=42)
shp_clf.fit(X_train, y_train)
# Get the number of extracted shapelets, the (minimal) distances from
# each of the timeseries to each of the shapelets, and the corresponding
# locations (index) where the minimal distance was found
n_shapelets = sum(shapelet_sizes.values())
distances = shp_clf.transform(X_train)
predicted_locations = shp_clf.locate(X_train)
plt.figure()
plt.title("Example locations of shapelet matches "
"({} shapelets extracted)".format(n_shapelets))
# Plot the test timeseries with the best matches with the shapelets
test_ts_id = numpy.argmin(numpy.sum(distances, axis=1))
plt.plot(X_train[test_ts_id].ravel())
# Plot the shapelets on their best-matching locations
for idx_shp, shp in enumerate(shp_clf.shapelets_):
t0 = predicted_locations[test_ts_id, idx_shp]
plt.plot(numpy.arange(t0, t0 + len(shp)), shp, linewidth=2)