My little research from a number of ANN-related articles reveal that authors are able to extract training, validation, testing and all data. I believe these are what MATLAB uses to generate the performance and R-squared values for the data sets respectively. Some authors calculate AAPE values for the mentioned sets of data.
These are my questions:
1. How can I extract these sets of data from my network? How are the AAPE values obtained? Are they calculated after the various values of the data set , say the testing data set are extracted?
2. I calculated the R-squared of the predicted values from my model. However, I find that it is different from what is displayed in the figure generated in the MATLAB graphs. Any reasons for the differences? How can i solve this?
Thanks.
Hi Joseph,
Kindly see the answers below.
1. How can I extract these sets of data from my network? How are the AAPE values obtained? Are they calculated after the various values of the data set , say the testing data set are extracted?
How can I extract the training, validation and testing data from my developed network?.
You could use the "divideParam" to do divide the data into training, validation, and test. Any evaluation metric can be computed once learning process is accomplished.
2. I calculated the R-squared of the predicted values from my model. However, I find that it is different from what is displayed in the figure generated in the MATLAB graphs. Any reasons for the differences? How can i solve this?
How can I extract the training, validation and testing data from my developed network?
This is because you've selected the wrong variables to correlate. AFAIK, the default plot shows R (not R-squared). However, once you compute R correctly, square the resulting value to get R^2.
HTH.
Samer
Hi Samer,
I am grateful for your rapid response.
1. I used the "divideParam" to divide the data into training, validation, and test. Training the network has been accomplished. I have even extracted the predicted values.
While exploring I found how MATLAB distributes the input data for the the purposes assigned. However, is there a way to extract predicted values categorised as training, validation and testing as has been achieved by some researchers?
2. You are right and thanks for clarifying it. The coefficient calculated by MATLAB is R not R-squared. I also confirmed squaring the R to get the R-squared for my work.
Thanks again, Samer.
Hi dear
Please see the supplementary file of my article
Article Performance and modeling of UV/Persulfate/Ce(IV) process as ...
I presented a m.file in details that can be useful for you
- Badges
- Science topic