3/06/2017

Focus more on theoretical function

Today, I tried several algorithms and adjust some parameters of ANN to improve the model. But R2 still did not improve much (about 0.5<R<0.6).

Summary:

 The following may be the best model I get today.
The first are comparisons of predictions and targets.


The second is all 3042 first parameters.
The third is log10-scale plot of y-axis of 3042 first parameters.
 The fourth is the list of algorithms.

Tomorrow, I will try to look into these algorithms to find the best one for my research.
Posted by at

15 comments:

  1. siddharth misraMonday, March 6, 2017 at 11:50:00 PM CST

    after trying the new algorithm

    , cluster the six parameters - means , deviations, and amplitudes - into groups.

    After clsutering into groups... try predicting the groups for each depth.

    meet me tomorrow morning to discuss further

    ReplyDelete
  2. siddharth misraMonday, March 6, 2017 at 11:52:00 PM CST

    print as six subplots - 0 to 500, 501 to 1000, 1001 to 1500, etc...this will help me follow the trends

    print y axis as log and another log axis as linear.

    ReplyDelete
  3. siddharth misraTuesday, March 7, 2017 at 12:07:00 AM CST

    can you identify the depths that are giving R>0.9 and higher accuracy in prediction?

    What is special about those depths with R>0.9?

    ReplyDelete
    Replies
    1. UnknownTuesday, March 7, 2017 at 10:34:00 AM CST

      Ok, I will try to find them out.

      Delete
    2. UnknownTuesday, March 7, 2017 at 11:05:00 AM CST

      R is a value for the total performance of the prediction. the calculation of R uses all 3042 original and 3042 predicted parameter values. So R cannot represent some depths.

      Delete
  4. siddharth misraTuesday, March 7, 2017 at 12:08:00 AM CST

    what depths are resulting in bad accuracy?

    What happens if you remove the depths with R<0.6

    ReplyDelete
    Replies
    1. UnknownTuesday, March 7, 2017 at 10:36:00 AM CST

      I think there are so many depths with R<0.6, maybe more than half, so it may be unrealistic to remove all of them out.

      Delete
    2. UnknownTuesday, March 7, 2017 at 11:09:00 AM CST

      although i cannot divide depths with higher and lower R as i said above, i may divide them with better and worse performance.

      Delete
  5. siddharth misraTuesday, March 7, 2017 at 12:10:00 AM CST

    what is on y-axis? please label all your axis with units

    ReplyDelete
    Replies
    1. UnknownTuesday, March 7, 2017 at 10:36:00 AM CST

      Ok, I forgot yesterday.

      Delete
    2. UnknownTuesday, March 7, 2017 at 10:39:00 AM CST

      Sorry, the y-axis is the parameter number. so it is not very meaningful to give it a unit. it is better to just let it be a number.

      Delete
  6. siddharth misraTuesday, March 7, 2017 at 12:11:00 AM CST

    what happens when you remove the logs that are not significantly contributing to the predictions? Some logs are causing the inaccuracy in predictions? Can you identify those logs and remove them?

    ReplyDelete
  7. siddharth misraTuesday, March 7, 2017 at 11:14:00 AM CST

    NMR sees pore structure...proe strcuture also depends on minerals -- carboantes clays etc.... a log that that is least sensitive to pore structure should be remvoed

    ReplyDelete

Subscribe to: Post Comments (Atom)