Skip to content

[ENH] Optimize ElasticEnsemble _fit to avoid redundant cross-validation #3109

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 8 commits into
base: main
Choose a base branch
from
Open

[ENH] Optimize ElasticEnsemble _fit to avoid redundant cross-validation #3109

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
fe686d7
feat: Optimize ElasticEnsemble _fit to avoid redundant CV
Nithurshen Nov 16, 2025
5ab5eb1
Merge branch 'main' into optimised-elastic-ensemble-grid-search
Nithurshen Nov 16, 2025
14aaac8
feat: Optimize ElasticEnsemble _fit to avoid redundant CV
Nithurshen Nov 16, 2025
bebc84b
Fixing modifications
Nithurshen Nov 16, 2025
a7b359b
Fixing modifications
Nithurshen Nov 16, 2025
5947193
Fixing modifications
Nithurshen Nov 16, 2025
1fcde54
Merge branch 'main' into optimised-elastic-ensemble-grid-search
Nithurshen Nov 16, 2025
0c1e24c
Adding Changelog
Nithurshen Nov 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
174 changes: 121 additions & 53 deletions aeon/classification/distance_based/_elastic_ensemble.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -251,62 +251,130 @@ def _fit(self, X, y):
f"Currently evaluating {self._distance_measures[dm]}"
)

# If 100 parameter options are being considered per measure,
# use a GridSearchCV
if self.proportion_of_param_options == 1:
grid = GridSearchCV(
estimator=KNeighborsTimeSeriesClassifier(
distance=this_measure, n_neighbors=1
),
param_grid=ElasticEnsemble._get_100_param_options(
self._distance_measures[dm], X
),
cv=LeaveOneOut(),
scoring="accuracy",
n_jobs=self._n_jobs,
verbose=self.verbose,
)
grid.fit(param_train_to_use, param_train_y)
best_distance_params = None
acc = 1.0 # Default for majority vote

# Optimized path:
# If we use all training data for param finding AND
# we are not using majority_vote (which needs weighting),
# we can combine the param search and accuracy estimation
# into a single loop to avoid the redundant CV pass.
if self.proportion_train_in_param_finding == 1.0 and not self.majority_vote:
if self.verbose > 0:
print( # noqa: T201
f"Using optimized manual CV path for "
f"{self._distance_measures[dm]}"
)

# Else, used RandomizedSearchCV to randomly sample parameter
# options for each measure
else:
grid = RandomizedSearchCV(
estimator=KNeighborsTimeSeriesClassifier(
distance=this_measure, n_neighbors=1
),
param_distributions=ElasticEnsemble._get_100_param_options(
self._distance_measures[dm], X
),
n_iter=math.ceil(100 * self.proportion_of_param_options),
cv=LeaveOneOut(),
scoring="accuracy",
n_jobs=self._n_jobs,
random_state=rand,
verbose=self.verbose,
param_grid = ElasticEnsemble._get_100_param_options(
self._distance_measures[dm], X
)
grid.fit(param_train_to_use, param_train_y)

if self.majority_vote:
acc = 1
# once the best parameter option has been estimated on the
# training data, perform a final pass with this parameter option
# to get the individual predictions with cross_cal_predict (
# Note: optimisation potentially possible here if a GridSearchCV
# was used previously. TO-DO: determine how to extract
# predictions for the best param option from GridSearchCV)
all_params = param_grid["distance_params"]

# Handle randomized search (proportion_of_param_options)
if self.proportion_of_param_options < 1:
n_iter = math.ceil(
len(all_params) * self.proportion_of_param_options
)
# Use a copy and shuffle to mimic RandomizedSearchCV
params_to_search = all_params.copy()
rand.shuffle(params_to_search)
params_to_search = params_to_search[:n_iter]
else:
params_to_search = all_params

best_acc = -1.0
best_distance_params = None

for params in params_to_search:
model = KNeighborsTimeSeriesClassifier(
n_neighbors=1,
distance=this_measure,
distance_params=params,
n_jobs=self._n_jobs,
)
# This CV is run on the FULL training set
preds = cross_val_predict(
model, full_train_to_use, y, cv=LeaveOneOut()
)
current_acc = accuracy_score(y, preds)

if current_acc > best_acc:
best_acc = current_acc
best_distance_params = params

acc = best_acc # Set the final accuracy for weighting

# Standard (original) path:
# This path is used if:
# 1. We are using a SUBSET of data for param finding, OR
# 2. We ARE using majority_vote.
else:
best_model = KNeighborsTimeSeriesClassifier(
n_neighbors=1,
distance=this_measure,
distance_params=grid.best_params_["distance_params"],
n_jobs=self._n_jobs,
)
preds = cross_val_predict(
best_model, full_train_to_use, y, cv=LeaveOneOut()
)
acc = accuracy_score(y, preds)
if self.verbose > 0:
print( # noqa: T201
f"Using standard GridSearchCV/RandomizedSearchCV "
f"path for {self._distance_measures[dm]}"
)

# If 100 parameter options are being considered per measure,
# use a GridSearchCV
if self.proportion_of_param_options == 1:
grid = GridSearchCV(
estimator=KNeighborsTimeSeriesClassifier(
distance=this_measure, n_neighbors=1
),
param_grid=ElasticEnsemble._get_100_param_options(
self._distance_measures[dm], X
),
cv=LeaveOneOut(),
scoring="accuracy",
n_jobs=self._n_jobs,
verbose=self.verbose,
)
grid.fit(param_train_to_use, param_train_y)

# Else, used RandomizedSearchCV to randomly sample parameter
# options for each measure
else:
grid = RandomizedSearchCV(
estimator=KNeighborsTimeSeriesClassifier(
distance=this_measure, n_neighbors=1
),
param_distributions=ElasticEnsemble._get_100_param_options(
self._distance_measures[dm], X
),
n_iter=math.ceil(100 * self.proportion_of_param_options),
cv=LeaveOneOut(),
scoring="accuracy",
n_jobs=self._n_jobs,
random_state=rand,
verbose=self.verbose,
)
grid.fit(param_train_to_use, param_train_y)

best_distance_params = grid.best_params_["distance_params"]

if self.majority_vote:
acc = 1
# once the best parameter option has been estimated on the
# training data, perform a final pass with this parameter option
# to get the individual predictions with cross_cal_predict (
# Note: optimisation potentially possible here if a GridSearchCV
# was used previously. TO-DO: determine how to extract
# predictions for the best param option from GridSearchCV)
else:
best_model = KNeighborsTimeSeriesClassifier(
n_neighbors=1,
distance=this_measure,
distance_params=grid.best_params_["distance_params"],
n_jobs=self._n_jobs,
)
preds = cross_val_predict(
best_model, full_train_to_use, y, cv=LeaveOneOut()
)
acc = accuracy_score(y, preds)

# Common code for both paths
if self.verbose > 0:
print( # noqa: T201
f"Training acc for {self._distance_measures[dm]}: {acc}"
Expand All @@ -317,7 +385,7 @@ def _fit(self, X, y):
best_model = KNeighborsTimeSeriesClassifier(
n_neighbors=1,
distance=this_measure,
distance_params=grid.best_params_["distance_params"],
distance_params=best_distance_params,
)
best_model.fit(full_train_to_use, y)
end_build_time = time.time()
Expand Down
1 change: 1 addition & 0 deletions docs/changelogs/v1.3.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,7 @@ September 2025

### Enhancements

- [ENH] Optimize `ElasticEnsemble` `_fit` to avoid redundant cross-validation** ({pr}`3109`) {user}`Nithurshen`
- [ENH] Improvements to ST transformer and classifier ({pr}`2968`) {user}`MatthewMiddlehurst`
- [ENH] KNN n_jobs and updated kneighbours method ({pr}`2578`) {user}`chrisholder`
- [ENH] Refactor signature code ({pr}`2943`) {user}`TonyBagnall`
Expand Down