Slides shown at experts building meeting Aug. 2nd, 2023.

William Luszczak  1:30 PM

This is the directory with the current PUEO antenna data files: https://github.com/PUEOCollaboration/pueoSim/tree/main/data/antennas. Each type of antenna will need several files:

• vv_0 and hh_0 describe the on axis v and h pol gain as a function of frequency. Gain is listed as a multiplicative factor (not in dB, if I'm remembering correctly)
• hv_0 and vh_0 describe the on axis cross polarization gain
• vv_az and hh_az describe the off-axis gain via multiplicative factors of the boresight gain. These are currently listed as a function of frequency for reference angles of 0,10,20,30,40,50,60,70,80, and 90 degrees, though we can always adjust the reference angles if needed.
• vv_el and hh_el are similar to above, but for gain as a function of elevation angle instead of azimuthal angle.

So for example, the total vpol gain in a particular frequency bin for a signal incident at azimuthal angle az and elevation angle el would be vv_0(f)*vv_az(f, az)*vv_el(f, el)

New

1:31

We can also potentially adjust this if it would be more convenient for the GENETIS people to output antenna information in a different format (or at different reference angles or whatever). The important thing is that the boresight h, v, and cross pol gains are all defined, as well as the off-axis response (as a function of azimuth and elevation)

Mostly just wrote more on the paper in the Genetic Algorithm section. I added some citation that we used in ICRC but there are still more places that should have citations.

I might check tonight when OSC is back up to try to push in more updates to the loop because I wanna get Evelyn and Ryan started on running the loop. Putting in those fixes is a big priority because we want to be able to correct the potential issue with XF simulation folders being overwritten and thus uan data not corresponding to the write individuals. The two big things for me in the loop are getting the simulation data to save correctly (and also putting that in the database) and testing that we can replicate results using the specific seed. I'll probably only focus on loop stuff tomorrow.

Got up before OSC was down to check progress of overnight run, it seems to have worked but I noticed a problem with the database that it wasn't writing to it probably due to a permissions issue but I would have to run another time to see. Shouldn't have affected data but just the use of the database. Run got up to 8 generations with non-zero fitness score which is positive and seems to have fixed the error we originally encountered. Talked to Eliot about pointers and possible errors but was unable to look at the specific error because it is on OSC.

Read about pointers and vectors in C++. Talked to Alex P a bit, and have some ideas of things to change to get the GA running. Began reading about antennas. Mostly a down day due to OSC being down.

Will implement changes to GA and continue familiarizing myself with how XF reads these values. 

Currently running the loop without the database to change the fix Alex and I made for XF to properly place data in the simulation folders. It looks to us like this was a problem that needed to be fixed. I'm running with the specific seed set in the roulette algorithm so we can replicate the results. I showed Evelyn and Ryan how to run the code today. Ryan was able to get a Pitzer Desktop, so until we can figure out how to get his X11 forwarding to work for XF, that will be a temporary fix. The only drawback is that Pitzer desktops require a GPU, so wait times can be high. I started this run by having him screenshare on zoom and then guiding him through running the loop. We got through the submission of the first batch of XF jobs (and also found some errors, so we had a chance to demonstrate fixing those).

My goal is to get several generations of this run done so that I have enough data to see if my fix worked. If so, I'm going to implement it in the database version and we'll perform a test with that. I'm meeting with Evelyn and Ryan at 1:30 to discuss the loop again, so I'll try to have a run going near where we left off today so they can see the rest. Once we know that we've fixed this XF problem we should be ready to fully implement the database by merging with the master branch.

Continued working on the erros with the roulette alg. I have gotten the code to a running state, however the data is not outputting correctly. I believe this is an issue within the roulette function or dataread function. 

I will further investigate roulette and dataread function and meet with Julie to do so. 

Alex M. Plans on helping me finish learning how to run the loops.

Continued the run testing the XF fixes with Alex P. We think our fixes are working, but we want to get a few more generations through. We checked the uan files and saw different data, which we didn't see on the past runs, so this indicates the fix is working. Alex and I also showed Ryan and Evelyn more of the loop and how to interpret the plots being generated. I also put in the red-green color plots, but we need to get through the next generation to see if they work. 

I'm going to try to get through more generations tonight so we can see what other fixes we still need to implement. I'll continue looking at any bugs that arise tomorrow, but we're getting very close to being able to mergre branches.

Font Continued to work on erros within roulette algorithm. Met with Julie to discuss the issue we were still having. Soon after meeting, Leo and I got it to a near perfect state. There are only a couple genese that were unique that shouldn't have been. This is likely a small error that I will look into tomorrow.

I plan to finalize the roulette alg to accept 2 chromosomes with 3 genes a piece. These will have shared radius and theta but unique lengths.

Met with Alex P. and Amy to think about how we can make sure the data coming from AraSim makes sense (specifically, to figure out why we're getting effective volumes of 0 on some antennas). Amy suggested we use a function in AraSim that let's us simulate identical events so that we can compare the same events between two different antennas. I asked Jorge in the #general chat to see if he could point us to where to look in AraSim for that function.

Met with Amy and Julie about people interested in joining the group and what upcoming project we have. We also talked about Karoo and Kai's thesis and how we can use machine learning to classify noise in the Anita data.

Alex P.

Today we finalized the GA. It now works with 2 chromosomes, where only length varies across chromosomes. In addition we got the tournament selection working (we don’t think it was working for even 1 chromosome prior). It is ready to begin being implemented tomorrow. We also met with Julie and others to discuss git and related topics.

Tomorrow we will implement the GA and begin editing Xmacros to 1) tell XF the new antenna type, etc and 2) have it read the new 2 chromosome data properly. This will likely take multiple days.

I wasn't available Thursday or Friday morning.

Alex P.

Spoke with Evelyn and Ryan about the paperclip antennas. They found the github repository with the files and a README so they're getting started on learning about the algorithm and testing out the scripts. Once they play with it and can get it running they'll start running it with different parameters for the selection methods (roulette vs tournament).

Also worked with Alex P. on figuring out the issue with the effective volumes coming out of AraSim. We have data from the eventReadIn function in AraSim for the small antennas and we don't see differences between the gains in these antennas and the actual bicone, but we still need to look at more of the quantities to compare that Amy told us about last week. We also came up with a new plot that might be interesting to look at. We think the plots look suspicious--I think it will be worthwhile to try rerunning with the same antennas but using a smaller grid spacing to see if the shape of the plots drastically changes.

Alex P.

Worked with Alex on trying to compare bicones to the ARA bicone. Met with Ben to give him some things to look over and told him to come to the meeting tomorrow at 1.

Alex P.

Talked with Alex about results from a run of small antennas where we decreased the grid spacing. The average gain plots look different, but I don't know yet if it's definitely resolved the problem because we're still getting a decent spread on the effective volumes. Helped Eliot and Leo with a git issue and talked to Evelyn and Ryan about paperclip. They were able to run and made some images, which is a good first step. I think they're going to start on making roulette an option since paperclip only uses tournament. 

Julie gave me some more details about using Karoo with ANITA data. 

Alex P.

Forgot to do updates this week so this is entire week. We have the loop running with two chromosomes up through Part C. ie we edited xmacros and other related parts to run simulation/XF with an asym bicone. Part D shouldn't need any changes, and Part E/F will need some. 

Friday/through the weekend we will each do git on codecademy and next week we will implement changes in the fitness function and plottinf(Part E/F)

Helped Alex P try to figure out how to get data Amy was asking us to get for comparing antennas in AraSim. We've been stuck trying to figure out where all of the data is printed (there are a bunch of .cc files and quantities we aren't familiar with--for example, vmmhz in Report.cc sounds like the V/Mhz with the effective heights folded in, but we aren't sure). 

I worked on some things in the paper that were mentioned in the minutes from last week's meeting, but I didn't see Amy's comments til the GENETIS meeting. 

Monday morning Amy might stop by to talk to Alex and I about AraSim so we can try to find some of the info we haven't been able to print out. We'll also watch out for anything she posts in slack about it.

We'll present an outline of the paper at 12 on monday to the Ara group meeting to see if it needs to be an ARA paper or not. 

I'm also going to keep looking for how to use the automatic grid spacing (and I'm planning on enlisting Eliot and Leo's help in being able to look at models in XF to make sure XF is designing antennas correctly).

Alex P.

Worked with Alex on finding the variables Amy suggested to us in AraSim. We found the Efield and max peak voltage variables and printed them, comparing three antennas: a small one with normal veffective (nonzero), a small on with 0 veffective, and the actual bicone. The electic field was the same for all antennas, but the peak voltages were all very low for the antenna with 0 veffective. Actually, after reinvestigating, it looks like the maximum peak voltages *were* the same between all the antennas.

I've been looking into using the automatic grid spacing in XF. I think I need to get a better understanding of XF, so I started playing with it to design things in the GUI. I want to be able to make a model using the xmacros that I can look at in the GUI so I can understand all of the functions being used. Since we've checked AraSim pretty extensively, we think that the issue is with XF. I think the grid spacing might be allowing small antennas to appear better than they are.

I think to investigate if the problem is in XF we should do a run of the same antenna with different grid spacing and look at the average gain plots. A preliminary test could be to try modeling a small antenna with a very large grid size--if that gives a very high gain, then it would suggest that the issue is with grid spacing being too large for small antennas.

I think I'll also keep playing with Karoo. I've been reviewing some ML material to refamiliarize myself and see if I can write a few simple programs. I got Karoo running on OSC last week but I don't really understand the outputs (using the practice dataset).

Alex P.

Worked with Alex P on looking for the problem in AraSim. We looked in Report.cc where Amy suggested there was an inequality involving a variable called "Full_window" being compared to the trigger threshold. We followed this down to another function in Report.cc where Full_window is used and compared fdiode_real between individuals, which was identical. We then tried printing V_Total_forconvlv, which was different between indviduals, yielding either all 0s or -nans in the individual with 0 effective volume. 

Helped Ryan some more with working on the roulette algorithm in paperclip. We're going to take it step by step and start with getting a hang of the code with print statements and modifications to see how things work and adding in a function for the roulette algorithm that will simply select the individuals. Once we have that working, we can move onto the next step of properly implementing the roulette algorithm in the code in a manner similar to what we have in the bicone algorithm. 

I made some small edits in the paper, mostly just pulling some citations from the proceeding but I also think I fixed the format of the references so that the numbers look nicer.

Alex P and I implemented the changes we discussed with Amy to Detector.cc. We're currently running a test to see how the grid spacing affects the gains from XF. We started with small antennas using a grid spacing of 0.01 and we're cutting down by factors of 2 each generation (until we see that the gains stop changing, then we'll do something of a binary search between the last two grid sizes to search for the largest reasonable size).

I also played around with the XF scripting more. My goal is to try to be able to edit the scripts we use to do help us check our results, like save images and load models for the antennas. I can currently run our macro script in XF by loading it in with 1 individual (using a generationDNA.csv file I moved to my user space). I can view and play with the model and it saves an image of it when I run the script (though the image doesn't give much detail). Hopefully this will lead to better understanding XF and being able to check our work and use the progrid option.

Worked with Alex to run XF with smaller grid spacing. We're down to 0.001 cm, but below that we were getting long times on the XF GUI that made it seem like it was hanging. I also investigated XF from a Pitzer desktop to see how to use the progrid setting. I can get the scripts to load--I checked that the original one is indeed making a bicone and that Eliot and Leo's makes an asymmetric bicone.

Helped Ryan some more with the paper clip. We're taking it step by step, so the next thing he'll want to do is make sure he can get the function for the roulette algorithm to be called correctly in the paperclip script and have the correct information passed to it (like the fitness scores).

Worked with Aelx P on running the loop with decreasing grid spacing. I also worked in the XF GUI some more and figured out how to make 3D gain plots (per Cade's suggestion) and tried comparing different grid sizes for a single antenna with proGrid on.