finish the NMR paper

Today, I finished the NMR paper.

Summary:

I sent you by an email.

I did three tests today for selecting inputs:
1. delete 10 inversion logs: R2 dropped to 0.7521.
2. delete 12 conventional logs: R2 dropped to 0.7889.
3. delete AT90 and VPVS logs: R2 dropped to 0.8423.

For the first two cases, apparent drops are obtained. So I recommend not to delete either of them. For the third case, the least important two inputs are deleted and similar accuracy as the original one is obtained. But since we have analyzed the importance of inputs in the paper, I think there is no need to delete them in this paper. Actually, I remembered that we discussed about it before. First, there is a big accuracy drop without either conventional logs or inversion logs. Second, since we have all of these 22 logs (without flags), it is better to use them together to predict NMR t2 distribution. There is no need to delete 2 or 3 of them if we want to keep similar accuracy.
I did not use k-fold cross-validation method because 85% training data and 15% testing data have been selected already before training and testing models. Instead, I did parallel computing for dividing data and training models. Almost the same accuracy is obtained so that I did not change results and figures in the paper.

The changed paper can be seen in the email.

NMR sequential prediction

1. column d is prediction of every group in one ANN model separately.
2. column e is sequential prediction.

Column a is accuracy of predicting 64 bins together and measuring accuracy of every group while normalized using maximum and minimum values of 64 bins. Column b is the ranking result. Column c is the number of bins in every group.

From the results, we can see that the two methods that we discussed yesterday do not perform better than predicting 64 bins together. The reason may be:

1. 64 bins together represent the distribution of pore size.
2. some group just means some pore size, which cannot be closely correlated to those input logs. The concentration of pore size at every depth is different from each other.
3. what matters is the dominated pore size at every depth, which is different from each other.

I think the sequential prediction for NMR and DD logs are different. 8 DD logs are kind of independent of each other at different frequencies, each representing conductivity or permittivity of the reservoir. However, 64 bins and 8 groups at every depth represent pore size distribution together, which cannot be divided. I think that is why the two methods cannot predict NMR t2 distribution separately with high accuracy.

Tomorrow, I will select inputs and do k-fold cross validation.

prepare for the defense

Today, I prepared for the defense.

Summary:

I recited the contents for slides. I also changed some slides from contents to figures.

Tomorrow, I will continue to prepare for the defense.

finish the ppt

Today, I finish the ppt for defense.

Summary:

I prepare for contents in every slide.

Tomorrow, I will deal with your suggestions for NMR paper.

prepare for the presentation

Today, I prepare for the presentation.

Summary:

I think about what to say in every slide.

Tomorrow, I will continue to prepare for it.

finish the draft of ppt

Today, I finish the draft of ppt.

Summary:

I just finish all contents in the ppt.

Tomorrow, I will start to prepare the presentation and improve the ppt.

make ppt for defense

Today, I made PPT for defense.

Summary:

I finish about 50%.

Tomorrow, I will try to finish all of them.

improve the thesis

Today, I continue to improve the thesis.

Summary:

I may finish about 70%.

Tomorrow, I will try to finish the thesis.

improve the paper

Today, I continue to improve the paper.

Summary:

I may finish about 50%.

Tomorrow, I will continue to improve the paper.

improve the second version of thesis

Today,  I started to improve the second version of thesis.

Summary:

I may finish 20%.

Tomorrow, I will continue to improve it.

k-mean method

Today, I compared clustering methods.

Summary:

I compared those clustering methods before and decided to use k-mean method to do clustering.
I am now looking for the best method to optimize the number of clusters.

Tomorrow, I will continue to do research on it.

select figures

Today, I selected figures for defense.

Summary:

I sent you by an email.

Tomorrow, I will start to write the thesis.

analyze 15 logs

Today, I analyze 15 logs

Summary:

The result is also not good for all models.

Tomorrow, I will try different number of logs and start to write the thesis.

add SOM

Today, I added SOM.

Summary:

For now, I just cluster sonic log, namely DTC and DTS because it is visible. From the scatter plot, we can see that sonic log is better to be divided into 2 or 3 clusters, which are close to prediction performance levels. As a result, the second and third figures are plotted after comparing prediction performance with 4 clustering models. However, some of them are close to prediction results while others are not.
I think that is the best that I can do if only sonic log is included for clustering.

Sonic plot:

Two clusters:

Three clusters:

Tomorrow, I will include some inputs into these models to see if more clusters can be determined and if results could be improved.

clustering models

Today, I built three clustering models for sonic log.

Summary:

Three models are:
1. k-means clustering
2. density-based spatial clustering of applications with noise (DBSCAN)
3. expectation maximization algorithm for gaussian mixture model (EM GMM)
The following are results of two clusters and three clusters.

Two clusters

Three clusters

From two figures, I think that two clusters performs better than three clusters. For k-means, it includes medium and poor clusters together and for DBSCAN and EM GMM, they include good and medium clusters together.

Tomorrow, I will try to improve the results.

write the outline of sonic log paper

Today, I wrote the outline of sonic log paper.

Summary:

I will finish it when it is done today.

Tomorrow, I will start to write the paper.

finish the Geophysics paper and start the sonic paper

Today, I finished the Geophysics paper and started the sonic paper.

Summary:

I sent you by an email.

Next week, I will continue to work on the sonic paper.

change the comments

Today, I changed the comments.

Summary:

I am now writing the modification part.

Tomorrow, I will finish it.

finish the improvement

Today, I finished the improvement of the geophysics paper.

Summary:

I sent you by an email.

Tomorrow, I will improve some contents of the paper.

continue to improve the paper

Today, I continued to improve the paper.

Summary:

I cannot finish it today. It may take another 2 or 3 days. I will try to finish it this week.

Tomorrow, I will continue to improve the paper.

finish the GJI paper

Today, I finished changing the format for GJI paper.

Summary:

I changed the format for GJI paper and continue to improve the geophysics paper.

Next week, I will continue to improve the geophysics paper.

continue to improve the paper

Today, I continued to improve the paper.

Summary:

I go on with another 3 comments.

Tomorrow, I will change the format of the rejected paper.

deal with the comments of the second reviewer

Today, I dealt with the comments of the second reviewer.

Summary:

There are about 20 comments from the second reviewer. I will deal with them one by one.

Tomorrow, I will continue to deal with them.

improve the geophysics paper

Today, I continued to improve the geophysics paper.

Summary:

I have finished answering comments from the first reviewer.

Tomorrow: I will start to deal with the second reviewer.

start to improve the geophysics paper

Today, I started to improve the geophysics paper.

Summary:

I marked all comments in four colors.
I started to deal with them one by one.
I will discuss with you about them if I have problems.

Tomorrow: I will continue to improve the paper.

finish the NMR paper

Today, I finished the NMR paper.

Summary:

1. I read materials of feature selection. There should be no problem of my method to evaluate the importance of inputs in the ANN model.
2. There are many models to rank the importance or select a subset of input data. However, they are all targeted for specific models. For example, SVM-RFE is for SVM model. Relief-F is for regression model. As a result, ranking importance using these models will not show the actual importance in the ANN model.
3. So I recommend to rank importance in the ANN model by using the original method.

Next week, I may start for the geophysics paper.

change the wrong figure

Today, I changed the wrong figure.

Summary:

The day before yesterday, I plotted the importance result. But there are two problems.
1. I deleted inputs one by one without using 0 to replace them.
2. The accuracy drop was not divided by original accuracy.
I fixed them and plot the following figure, which should be no problem.

Tomorrow, I will continue to look for methods to rank importance and give you the final version of the paper.

rank importance of inputs

Today, I used other ways to rank importance of inputs.

Summary:

Feature selection is one method to decrease dimension of inputs while keeping as many information of inputs as possible. It select a subset of input for model construction. As a result, it is not for ranking importance of inputs.

I found another algorithm for ranking importance, which is called ReliefF.
The following is the result:

The ranking is different from what is calculated by deleting inputs one by one. As searched online, most people analyze importance by deleting inputs one by one. So I think that the original method is more trustworthy than the ReliefF algorithm.

Tomorrow, we can discuss about the result and I plan to go on the next task, which is find clustering method for sonic log prediction model.

finish accuracy drop analysis

Today, I finished accuracy drop analysis.

Summary:

The following is the result.

Tomorrow, I will try to use other ways to rank the importance of inputs.

coding for ranking inputs

Today, I coded for ranking inputs.

Summary:

I have not finished it yet. I will try to finish it tomorrow.
But my concern is that: since I have 27 inputs, deleting any one of them will not see apparent accuracy drop. So the ranking may not be feasible.

Tomorrow, I will try to run and see the results.

evaluation of GMM application in accuracy analysis

Today, I did evaluation of GMM application in accuracy analysis.

Summary:

I have not coded for the whole process, so I did not know whether it will work. But we can have a try if you want.

Tomorrow: I will continue to read some materials about GMM.

GMM may be for fitting problems.

Today, I learned about how to build a GMM in mATLAB.

Summary:

I knew how to use the GMM to fit data, but it may not be able for prediction.

GMM is used to fit data using several gaussian distributions. People can obtain parameters of these gaussian distributions from raw data, but it is about the fitting problem.
For now, I did not see any application of GMM similar to our problem.

Tomorrow, maybe we can have a talk about GMM.

continue to learn about the model

Today, I continued to learn about the model.

Summary:

I am learning how to build the model in Matlab.

Tomorrow, I will continue to build the model.

develop the gaussian mixture model

Today, I learned about the gaussian mixture model.

Summary:

I started to build the gaussian mixture model.

Tomorrow, I will continue to develop the model.

finish improving ATCE paper proposal

Today, I finished improving ATCE paper proposal.

Summary:

I sent you by an email.

Next week, I will start to work on sonic log paper.

improve two abstracts

Today, I improved two abstracts.

Summary:

I am now working on the ATCE paper proposal and SPWLA abstract.

Tomorrow, I will try to finish them.

finish the TGRS paper and start to write ATCE paper proposal

Today, I finished the TGRS paper and started to write ATCE paper proposal.

Summary:

I am now writing the ATCE paper proposal on sonic logs.

Tomorrow, I will try to finish the paper proposal.

continue to improve the paper

Today, I continued to improve the TGRS paper.

Summary:

I am now working on improving the sensitivity to noise in data.

Tomorrow, I will try to finish the TGRS paper.

continue to improve the TGRS paper

Today, I continued to improve the TGRS paper.

Summary:

I am now coding for re-ranking for all inputs to decide the number of inputs kept when accuracy drop does not exceed 10%.

Tomorrow, I will continue to improve the TGRS paper.

continue to improve the TGRS paper

Today, I continued to improve the TGRS paper.

1.22-1.26
finish the TGRS paper, submit the abstract for SPWLA

1.29-2.2
build the gaussian mixture model and compare it with other models

2.5-2.9
test the best model in another well

2.12-2.16
write the draft of sonic log paper

2.19-2.23
improve and finish the sonic log paper

Next week, I will continue to improve the TGRS paper.

finish the PPT and start to work on TGRS paper

Today, I finished the PPT and started to work in TGRS paper.

Summary:

I will send the PPT to Dr. Deepak and Dr. Wu tomorrow morning.
I will improve the abstract for SPWLA after you review it.

Tomorrow, I will continue to work on the TGRS paper.

finish the abstract for SPWLA

Today, I finished the abstract for SPWLA.

Summary:

I sent you by an email.

Tomorrow, I will make the ppt and improve the abstract.

finish checking the innovation and start to write the abstract

Today, I finished checking the innovation of the method in the second paper and start to write the abstract for SPWLA.

Summary:

I did literature review and did not find any paper using the same method as mine. I think it is innovative.

I start to write the abstract for SPWLA. I will try to finish it tomorrow and send it to you.

Tomorrow, I will continue to write the abstract for SPWLA.