Example with financial data #195
Conversation
* add dependence to imbalanced_learn add example with financial data * print score remove dead code * add patent application number * [pre-commit.ci] auto fixes from pre-commit.com hooks * Update financial_data.py * Update financial_data.py Co-authored-by: fbarroso24 <fbarroso24@gmail.com> --------- Co-authored-by: Gregoire Cattan <gregoire.cattan@ibm.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: fbarroso24 <fbarroso24@gmail.com>
| ) | ||
|
|
||
| ############################################################################## | ||
| # Run evaluation |
There was a problem hiding this comment.
What are the non-quantum state-of-the-art methods for detecting financial fraud?
It would be good to add them in the comparison.
There was a problem hiding this comment.
Good idea! May be some decision tree/random forst. @fbarroso24 what do you think?
| pipe, | ||
| param_grid={ | ||
| "toepochs__n": [10, 20], | ||
| "xdawncovariances__nfilter": [1, 2], |
There was a problem hiding this comment.
You should test higher values for nfilter.
What is the number of features?
There was a problem hiding this comment.
Only three in this example. We could add more features, but the simulation time is quite long.
There was a problem hiding this comment.
With more features, classical pipeline would perform much better.
It seems unfair that time issues linked to quantum pipeline hamper the performance of classical one.
There was a problem hiding this comment.
Makes sense. I can try, and find a compromise for the pipeline afterwhat.
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
- improve description - add train/test split
- missing propagation of X_train/X_test changes
- display best csv_results without pandas
| score_qsvm = gs.best_estimator_.fit(X_train, y_train).score(X_test, y_test) | ||
|
|
||
| # Print the results | ||
| print(f"Classical: {score_svm} \nQuantum : {score_qsvm}") |
There was a problem hiding this comment.
Quantum pipeline gives a binary classification score of 0,5.
Flipping a coin would do the same thing... unless I missed something.
There was a problem hiding this comment.
No, this is weird.
In the first version, it was 100%... but with the same data.
I need to investigate this.
- switch to halvinggridsearchcv
|
@qbarthelemy I made another pass on the example. There are mainly two changes:
I also gave a tried to the halving grid search. It is quicker, but may be less accurate than the standard grid search.
|
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
This is an example based on a patent application exploiting RG+quantum for detecting fraudulent behavior.