Skip to content

Optimize debias model training by removing extra GridSearchCV refit and vectorizing RMSE blend-weight search. Fixes Issue Number #3907 #3908

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
Vibhanshu230 wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

Optimize debias model training by removing extra GridSearchCV refit and vectorizing RMSE blend-weight search. Fixes Issue Number #3907 #3908

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
14 changes: 6 additions & 8 deletions modules/assim.sequential/inst/python/pecan_debias/debias.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ def _fit_knn(X, y):
grid = GridSearchCV(
pipe,
{'kneighborsregressor__n_neighbors': list(range(1, 31))},
cv=max(2, min(5, len(y))),
cv=min(5, max(2, len(y)//2)),
scoring='neg_root_mean_squared_error',
n_jobs=-1
)
Expand All @@ -35,13 +35,12 @@ def _fit_extratrees(X, y):
base = ExtraTreesRegressor(random_state=42, n_jobs=-1)
grid = GridSearchCV(
base, param_grid,
cv=max(2, min(5, len(y))),
cv=min(5, max(2, len(y)//2)),
scoring='neg_root_mean_squared_error',
n_jobs=-1
)
grid.fit(X, y)
tree = grid.best_estimator_
tree.fit(X, y)
return tree

def _fit_one(X, y):
Expand All @@ -51,11 +50,10 @@ def _fit_one(X, y):
knn_pred = knn.predict(X)
tree_pred = tree.predict(X)
weights = np.linspace(0, 1, 101)
best_w, best_rmse = 0.5, np.inf
for w in weights:
rmse = np.sqrt(mean_squared_error(y, w*knn_pred + (1-w)*tree_pred))
if rmse < best_rmse:
best_rmse, best_w = rmse, w
# replacing the loop based RMSE search over blend weights with a vectorised approach.
preds = weights[:, None]*knn_pred[None, :]+(1-weights[:, None])*tree_pred[None, :]
rmses = np.sqrt(((preds-y)**2).mean(axis=1))
best_w = weights[np.argmin(rmses)]
return knn, tree, float(best_w)

def train_full_model(name, X, y, feature_names=None):
Expand Down