Adapted the number of splits in shuffle split to increase speed in plot_learning_curve.py #21628
Adapted the number of splits in shuffle split to increase speed in plot_learning_curve.py #21628adrinjalali merged 5 commits intomainfrom unknown repository
Conversation
ogrisel
left a comment
There was a problem hiding this comment.
LGTM, could you please just propagate the ylim param to the last plot to make results comparable across models?
|
Please also pass
which can be very confusing. I will open a dedicated issue. Edit: actually the |
|
we could exclude the very fast runs from the example maybe? |
Added ``random_state=0`` to call of ``learning_curve``
|
@ogrisel Thanks for the input :) I added the |
As I explained in my edited comment, the
I want the y axis (score values) of both models to be on the same scale on the last row as they are on the first row. |
Maybe. Or we can keep the dataset large enough for the fastest model (NB) and subsample it only for the slowest model (SVC). Or we can just live with it. |
Added ``shuffle=True`` for ``random_state`` to make an impact
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
adrinjalali
left a comment
There was a problem hiding this comment.
The new plot doesn't look more odd than what we have. Merging this one, we can merge this one and leave the fix for a separate PR.
|
@adrinjalali Alright |
* Adapted the number of splits * Update plot_learning_curve.py * Update plot_learning_curve.py Added ``random_state=0`` to call of ``learning_curve`` * Update plot_learning_curve.py Added ``shuffle=True`` for ``random_state`` to make an impact * Update examples/model_selection/plot_learning_curve.py Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org> Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
* Adapted the number of splits * Update plot_learning_curve.py * Update plot_learning_curve.py Added ``random_state=0`` to call of ``learning_curve`` * Update plot_learning_curve.py Added ``shuffle=True`` for ``random_state`` to make an impact * Update examples/model_selection/plot_learning_curve.py Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org> Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
* Adapted the number of splits * Update plot_learning_curve.py * Update plot_learning_curve.py Added ``random_state=0`` to call of ``learning_curve`` * Update plot_learning_curve.py Added ``shuffle=True`` for ``random_state`` to make an impact * Update examples/model_selection/plot_learning_curve.py Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org> Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
* Adapted the number of splits * Update plot_learning_curve.py * Update plot_learning_curve.py Added ``random_state=0`` to call of ``learning_curve`` * Update plot_learning_curve.py Added ``shuffle=True`` for ``random_state`` to make an impact * Update examples/model_selection/plot_learning_curve.py Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org> Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
* Adapted the number of splits * Update plot_learning_curve.py * Update plot_learning_curve.py Added ``random_state=0`` to call of ``learning_curve`` * Update plot_learning_curve.py Added ``shuffle=True`` for ``random_state`` to make an impact * Update examples/model_selection/plot_learning_curve.py Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org> Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
#21598 @adrinjalali @sply88
Cut the number of splits in half twice. Didn't really change the outcome but time dropped significantly.
