Open ManikantaSanjay opened 2 years ago
Hi could you send the error please ? With which csv file do you have this problem ? And is this error breaking the code ? Because this error may happen during the hyperparameters tunning for some specific combinations of hyperparameters (when it leads to non positive semi-definite matrices due to precision limit ) but it shouldn't stop the search process.
Hi Gregory , This error isn’t breaking the code but I have a doubt if it is leading to the correct results coz the results that I have retrieved from it haven’t been good. I’m running it to predict PCA components for my time series data.
On Tue, 1 Mar 2022 at 7:08 PM, gregory scafarto @.***> wrote:
Hi could you send the error please ? With which csv file do you have this problem ? And is this error breaking the code ? Because this error may append during the hyperparameters tunning for some specific combination of hyperparameters (when it leads to non positive semi-definite matrices due to precision limit ) but it shouldn't stop the search process.
— Reply to this email directly, view it on GitHub https://github.com/gregoritoo/ABCDflow/issues/4#issuecomment-1055451418, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMZCYEKNIUF4J2Z7DPFKRUTU5YMT3ANCNFSM5MD2OCIQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you authored the thread.Message ID: @.***>
if you only have "Input matrix is not invertible. [Op:MatrixSolve]" results at one step then yes indeed the results may not be optimal. This is due to the enormous size of the search space with a lot of local minima (problem that the authors have highlighted in their original work). That's why the results are generally better with periodic signals. In order to improve the search, at each step , kernels are instantiate with the best hyperparameters form the previous step. If you are not time limited you can also increase the initialisation_restart parameter of the launch_analysis function, it will improve the quality of the results.
So I just need to increase that restart time to a larger value. Ok I’m facing this “input matrix is not invertible”. Is there a way to fix this?
On Tue, 1 Mar 2022 at 7:57 PM, gregory scafarto @.***> wrote:
if you only have "Input matrix is not invertible. [Op:MatrixSolve]" results at one step then yes indeed the results may not be optimal. This is due to the enormous size of the search space with a lot of local minima (problem that the authors have highlighted in their original work). That's why the results are generally better with periodic signals. In order to improve the search, at each step , kernels are instantiate with the best hyperparameters form the previous step. If you are not time limited you can also increase the initialisation_restart parameter of the launch_analysis function, it will improve the quality of the results.
— Reply to this email directly, view it on GitHub https://github.com/gregoritoo/ABCDflow/issues/4#issuecomment-1055501155, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMZCYEMGTIRMFOSLCYHLUODU5YSMTANCNFSM5MD2OCIQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you authored the thread.Message ID: @.***>
No but I'm currently working on adapting this process to the GPy librairy as it seems that their calculus are more stable.
Good to hear that. And also if you need any help with the repo development do let me know. I'm completely free on weekends. Keep the good work going. Thanks and Regards Manikanta Sanjay Veera
On Tue, 1 Mar, 2022, 8:07 PM gregory scafarto, @.***> wrote:
No but I'm currently working on adapting this process to the GPy librairy as it seems that their calculus are more stable.
— Reply to this email directly, view it on GitHub https://github.com/gregoritoo/ABCDflow/issues/4#issuecomment-1055510310, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMZCYEPOPQQPTP4WNYVEIC3U5YTSLANCNFSM5MD2OCIQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you authored the thread.Message ID: @.***>
@ManikantaSanjay the wrapper is updated, please feel free to try it and give feedback.
Thanks, I'll check it out.
On Wed, 16 Mar 2022 at 1:11 PM, gregory scafarto @.***> wrote:
@ManikantaSanjay https://github.com/ManikantaSanjay the wrapper is updated, please feel free to try it and give feedback.
— Reply to this email directly, view it on GitHub https://github.com/gregoritoo/ABCDflow/issues/4#issuecomment-1068825448, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMZCYEJWS6OKGV4O7RC3STTVAGGCBANCNFSM5MD2OCIQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you were mentioned.Message ID: @.***>
Input matrix is not invertible. [Op:MatrixSolve]. I'm receiving this error when I'm trying to run the code on the sample data mentioned in the repo.