There was a magnitude 6.6 earthquake just a few minutes ago. I am attaching photographs of the monitor feeds for reference here. Is there a standard protocol to be followed in this situation? I'm looking through the wiki now.

Further, the IMC seems to be misaligned and is not locking! As Koji has let me know, I really hope this is not too serious and can be fixed easily.

The quoted elog has figures which indicate that the network did not learn (train or generalize) on the used data. This is a scary thing as (in my experience) it indicates that something is fundamentally wrong with either the data or model and learning will not happen despite how hyperparameters are tuned. To check this, I ran the training experiment for nearly 25 hyperparameter settings (results here)with the old data and was able to successfully overfit the data. Why is this progress? Well, we know that we are on the right track and the task is to reduce overfitting. Whether, that will happen through more hyperparameter tuning, data collection or augmentation remains to be seen. See attachments for more details. 

Why is the fit so perfect at the start and bad later? Well, that's because the first 90% of the test data is  the training data I overfit to and the latter the validation data that the network has not generalized well to.

[Kruthi, Koji]

Koji came to the lab to align the IMC/IFO, but found the mirrors are dancing around. Kruthi told me that there was M7.1 EQ at Ridgecrest. Looks like there are aftershocks of this EQ going on. So we need to wait for an hour to start the alignment work.

ITMX and ETMX are stuck.

- ITM unstuck now
- IMC briefly locked at TEM00

A series of aftershocks came. I could unstick ITMX by turning on the damping during one of the aftershocks.
Between the aftershocks, MC1~3 were aligned to the previous dof values. This allowed the IMC flashing. Once I got the lock of a low order TEM mode, it was easy to recover the alignment to have a weak TEM00.
Now at least temporarily the full alignment of the IMC was recovered.

In fact, ETMX was not stuck until the M7.1 EQ today. After that it got stuck, but during the after shocks, all the OSEMs occasionally showed full swing of the light levels. So I believe the magnets are OK.

I modified the autolocker code I wrote to read from a .yaml configuration file instead of commandline arguements (that option still exists if one wishes to override what the .yaml file contains). I have pushed the code to github. I started reading about MCMC and will put up details of the remaining part of the work ASAP.

The beam splitter (BS1-1064-33-2037-45S) that is currently being used has an antireflection coating on the second surface and a wedge of less than 5 arcmin; yet it leads to ghosting as shown in the figure attached (courtesy: Thorlabs). I'm also attaching its spec sheet I dug up on internet for future reference.

I came across pellicle beamsplitters, that are primarily used to eliminate ghost images. Pellicle beamsplitters have a few microns thick nitrocellulose layer and superimpose the secondary reflection on the first one. Thus the ghost image is eliminated. 

Should we go ahead and order them? (https://www.thorlabs.com/newgrouppage9.cfm?objectgroup_id=898

https://www.edmundoptics.eu/c/beamsplitters/622/#28438=28438_s%3AUGVsbGljbGU1&27614=27614_d%3A%5B46.18%20TO%2077.73%5D)

I came across a paper (see reference) where they have used DAOPHOT, an astronomical software tool developed by NOAO, to study the point scatterers in LIGO test masses using images of varying exposure times. I'm going through the paper now. I think using this we can analyze the MC2 images and make some interesting observations.

Reference:  L.Glover et al., Optical scattering measurements and implications on thermal noise in Gravitational Wave detectors test-mass coatings Physics Letters A. 382. (2018)

After the two earthquakes, I collected some data by dithering the optic and recording the QPD readings. Today, I set up scripts to process the data and then train networks on this data. I have pushed all the code to github. I attempted to train a bunch of networks on the new data to test if the code was alright but realised quickly that, training on my local machine is not feasilble at all as training for 10 epochs took roughly 6 minutes. Therefore, I have placed a request for access to the cluster and am waiting for a reply. I will now set up a bunch of experiments to tune hyperparameters for this data and see what the results are.

Trainng networks with memory

I set up a network to handle input volumes (stacks of frames) instead of individual frames. It still uses 2D convolution and not 3D convolution. I am currently training on the new data. However, I was curious to see if it would provide any improved performance over the results I put up in the previous elog. After a bit of hyperparameter tuning, I did get some decent results which I have attached below. However, this is for Pooja's old data which makes them, ah, not so relevant. Also, this testing isn't truly representative because the test data isn't entirely new to the network. I am going to train this network on the new data now with the following objectives (in the following steps):

1. Train on data recorded at one frequency, generalize/ test on unseen data of the same frequency, large amplitude of motion
2. Train on data recorded at one frequency, generallize/ test on unseen data of a different frequency, large amplitude of motion
3. Train on data recorded at one frequency, generalize/ test on unseen data of  same/ different frequency, small amplitude of motion
4. Train on data at different frequencies and generalize/ test on data with a mixture of frequencies at small amplitudes - Gautam pointed out that the network would truly be superb (good?) if we can just predict the QPD output from the video of the beam spot when nothing is being shaken.

I hope this looks alright? Rana also suggested I try LSTMs today. I'll maybe code it up tomorrow. What I have in mind- A conv layer encoder, flatten, followed by an LSTM layer (why not plain RNNs? well LSTMs handle vanishing gradients, so why the hassle).

you have to use a BS with a larger wedge angle (5 arcmin ~ 1 mrad) so that the beams don't overlap on the camera

Today, Gautam keyed the C1PSL crate and we got to test my unstick.py code. It seems to be working fine. Remarks:

1. Gautam moved the unstick.py code to /opt/rtcds/caltech/c1/scripts/cds. Therefore, the steps to run this code are now:
   1. cd /opt/rtcds/caltech/c1/scripts/cds
   2. python unstick.py c1psl (for the c1psl machine)
2. There is now a sleepTime global variable in the code which defines the amount of delay between successive channel toggles. We set this to 1ms and it took the code around 3s to run.
3. Gautam was curious to see if this would work even if we set the sleepTime parameter to 0 but decided that that could be tested the next time something was keyed.
4. I still need to add the signal handling thing to this code.

Following this, we tested my PMC autolocker code. The code ran for about a minute before achieveing lock. Remarks:

1. Gautam moved my code (pmc_autolocker.py and autolocker_config.yaml) to /cvs/cds/rtcds/caltech/c1/scripts/PSL/PMC/ . Therefore, the steps to run this code are now:
   1. cd /cvs/cds/rtcds/caltech/c1/scripts/PSL/PMC/
   2. python pmc_autolocker.py (check code or use --help to see what the command line arguments do which is only for when you wanna override the details in the .yaml file)
2. Gautam suggested that I add some delay between succesive steps of DC output adjust so that it locks quickly. I'll do that ASAP. For now, it works.

The Y-arm ASS was tuned to be in a workable state. Basically, I followed Koji's recipe.

The SNR of the dither lines in the TRY and YARM control signals were checked - Attachment #1. The dither frequencies are marked with vertical dashed lines (can't figure out how to add 4 cursors in DTT so there's two in each row for a total of 4). A couple of days ago, when I was doing some preliminary checks, I found that the oscillator at 24.91 Hz caused a broadband increase in the TRY noise between DC and ~100 Hz. But today I saw no evidence of such behaviour. So I decided against changing the frequency.

The linearity of the demodulated error signals around the quadratic maxima of the TRY level was checked. I did not, however, investigate in detail the frequency-dependent offset Koji has reported in his elog. 

After this work, the TRY level is at 0.95. This is commensurate with the MC trans level being lower by ~7% relative to July 2018. Furthermore, the ASS servo is able to return to TRY~0.95 with a time-constant of ~5 seconds in response to misalignment of the cavity optics. After I investigate the X-arm ASS, I will reset the normalization for TRX and TRY.

Update 645pm: In the spirit of general IFO recovery, I re-centered the ITM and ETM oplev spots, and also the IR beam on the IPPOS QPD to mark the new input pointing alignment (the spot is slightly lower on the AS camera than what I remember). I then tweaked the XARM transmission to maximize it, and re-set the TransMon normalization. I edited the normalization script to comment out the normalizing of the TransMon QPD gains as the QPDs are in some kind of indeterminate state now. Attachment #2 shows the current status, you can also see the normalization being reset. LSC mode disabled for overnight.

Once the XARM ASS is also checked out, I propose moving back to locking the DRMI / PRFPMI configs. 

Last documented replacement in Nov 2018, so ~7 months, which I believe is par for the course. I am disconnecting its power supply cable.

There was no green light even though the EX NPRO was on. I checked the doubling oven temperature controller and found that its power cable was loose on the rear. I reconnected it, and now there is green light again. 

I received access today. After some incredible hassle, I was able to set up my repository and code on the remote system. Following this, Gautam wrote to Gabriele to ask him about which GPUs to use and if there was a previously set up environment I could directly use. Gabriele suggested that I use pcdev2 / pcdev3 / pcdev11 as they have good gpus. He also said that I could use source ~gabriele.vajente/virtualenv/bin/activate to use a virtualenv with tensorflow, numpy etc. preinstalled. However, I could not get that working, Therefore I created my own virtual environment with the necessary tensorflow, keras, scipy, numpy etc. libraries and suitable versions. On ssh-ing into the cluster, it can be activated using source /home/millind.vaddiraju/beamtrack/bin/activate. How do I know everything works? Well, I trained a network on it! With the new data. Attached (see attachment #1) is the prediction data for completely new test data. Yeah, its not great, but I got to observe the time it takes for the network to train for 50 epochs-

1. On pcdev5 CPU: one epoch took ~1500s which is roughly 25 minutes (see Attachment #2). Gautam suggested that I try to train my networks on Optimus. I think this evidence should be sufficient to decide against that idea.
2. On my GTX 1060: one epoch took ~30s. Which is 25 minutes (for 50 epochs) to train a network.
3. On pcdev11 GPU (Titan X I think): each epoch took ~16s which is a far more reasonable time.

Therefore, I will carry out all training only on this machine from now.

Note to self:

Steps to repeat what you did are:

1. ssh in to the cluster using ssh albert.einstein@ssh.ligo.org as described here.
2. activate virtualenv as descirbed above
3. navigate to code  and run it.

Arnaud and I moved one of the two spare TT suspensions from the south clean cabinet to the bake lab clean room. The main purpose was to inspect the contents of the packaging. According to the label, this suspension was cleaned to Class A standards, so we tried to be clean while handling it (frocks, gloves, masks etc). We found that the foil wrapping contained one suspension cage, with what looked like all the parts in a semi-assembled state. There were no OSEMs or electronics together with the suspension cage. Pictures were taken and uploaded to gPhoto. Arnaud is going to plan his tests, so in the meantime, this unit has been stored in Cabinet #6 in the bake lab cleanroom.

In fact the projector is still working. The lamp timer showed ~8200hrs. I just reset the timer, but not sure it was the cause of the shutdown. I also set the fan mode to be "High Altitude" to help cooling.

[koji, gautam]

We noticed that the PRM watchdog was tripping frequently. This is a period of enhanced seismic activity. The reason PRM in particular trips often is because the SIDE OSEM has 5x increased transimpedance. We implemented a workaround by modifying the watchdog tripping condition to scale the SD channel RMS by a factor of 0.2 (relative to the UL and LL channels). We restarted the modbus process on c1susaux and tested that the new logic works. Here is the relevant snippet of code:

The db file has a note about this as well so that future debuggers aren't mystified by a factor of 0.2.

I trained a bunch (around 25 or so - to tune hyperparameters) of networks today. They were all CNNs. They all produced garbage. I also looked at lstm networks with CNN encoders (see this very useful link) and gave some thought to what kind of architecture we want to use and how to go about programming it (in Keras, will use tensorflow if I feel like I need more control). I will code it up tomorrow after some thought and discussion. I am not sure if abandoning CNNs is the right thing to do or if I should continue probing this with more architectures and tuning attempts. Any thoughts?

Right now, after speaking to Stuart (ldas_admin) I've decided on coding up the LSTM thing and then running that on one machine while probing the CNN thing on another.

Update on 10 July, 2019: I'm attaching all the results of training here in case anyone is interested in the future.

I heard a popping sound in the control room; the projector lightbulb has blown out.

Arnaud has taken 1 TT suspension from the 40m clean lab to Downs for modal testing. Estimated time of return is tomorrow evening.

All suspension watchdogs were tripped ~90mins ago. I restored the damping. IMC is locked.

ITMX was stuck. I set it free. But notice that the UL Sensor RMS is higher than the other 4? I thought ITMY UL was problematic, but maybe ITMX has also failed, or maybe it's coincidence? Something for IFOtest to figure out I guess. I don't think there is a cable switch between ITMX/ITMY as when I move the ITMX actuators, the ITMX sensors respond and I can also see the optic moving on the camera.

Took me a while to figure out what's going on because we don't have the seis BLRMS - i moved the usual projector striptool traces to the TV screen for better diagnostic ability.

Update 16 July 1515: Even though the RMS is computed from the slow readback channels, for diagnosis, I looked at the spectra of the fast PD monitoring channels (i.e. *_SENSOR_*) for ITMX - looks like the increased UL RMS is coming from enhanced BR-mode coupling and not of any issues with the whitening switching (which seems to work as advertised, see Attachment #3, where the LL traces are meant to be representative of LL, LR, SD and UR channels).

Optical chopper borrowed from CryoLab to 40m

https://nodus.ligo.caltech.edu:8081/Cryo_Lab/2458

On Friday, I took images for different power outputs of LED. I calculated the calibration factor as explained in my previous elog (plots attached).

To estimate the uncertainity, I assumed an error of at most 20mV (due to stray lights or difference in orientation of GigE and photodiode) for the photodiode reading. Using the uncertainity in slope from the linear fit, I expect an uncertainity of maximum 4%. Note: I haven't accounted for the error in the responsivity value of the photodiode.

Johannes had reported CF as 0.0858E-15 W-sec/counts for 12 bit images, with measured a laser source. This value and the one I got are off by a factor of 25. Difference in the pixel formats and effect of coherence of the light used might be the possible reasons.

On Friday, Koji helped me find various components required for the scatterometer setup. Like he suggested, I'll first set it up on the SP table and try it out with an usual mirror. Later on, once I know it's working, I'll move the setup to the flow bench near the south arm and measure the BRDF of a spare end test mass.

I made some rough measurements, using the setup I had used for CCD calibration, to get an idea of how good of a Lambertian scatterer the white paper is. Following are the values I got:

Note: All the measurements are just rough ones and are prone to larger errors than estimated.

I also measured the transmittance of the white paper sample being used (it consists of 2 white papers wrapped together). It was around 0.002

I've set up network with a CNN encoder (front end) feeding into a single LSTM cell followed by the output layer (see attachment #1). The network requires significantly more memory than the previous ones. It takes around 30s for one epoch of training. Attached are the predicted yaw motion and the fft of the same. The FFT looks rather curious. I still haven't done any tuning and these are only the preliminary results.

Well, what about the previous conv nets?

What I did:

1. Extensive tuning - of learning rate, batch size, dropout ratio, input size using a grid search
2. Trained each network for 75 epochs and obtained weights, predicted motion and corresponding FFT, error etc.

What I observed:

1. Loss curves look okay, validation loss isn't going up, so I don't think overfitting is the issue
2. Training for over (even) 75 epochs seems to be pointless.

What I think is going wrong:

1. Input size- relatively large input size: 350 x 350. Here, the input image size seems to be 128 x 128.
2. Inadequate pre-processing.
   1. I have not applied any filters/blurs etc. to the frames.
   2. I have also not tried dimensionality reduction techniques such as PCA

What I will try now:

1. Collect new data: with smaller amplitudes and different frequencies
2. Tune the LSTM network for the data I have
3. Try new CNN architectures with more aggressive max pooling and fewer parameters
4. Ensembling the models (see this and this). Right now, I have multiple models trained either with same architecture and different hyperparameters or with different architectures. As a first pass, I intend to average the predictions of all the models and see if that improves performance.

Mode cleaner was not locked today. Koji came in and concluded that PSL had died. So we keyed it. Then we ran my unstick.py code. Mode cleaner is locked now.

[koji, gautam]

In addition to c1psl needing a reboot, megatron was un-ssh-able (although it was responding to ping). Clue was that the NPRO PZT control voltage was drifting a lot on the StripTool trace. Koji hard-rebooted the machine. Now IMC is locked, and FSS slow servo is also running.

There were several small/medium earthquakes in Ridgecrest and one medium one in Blackhawk CA at about 2000 UTC (i.e. ~ 2 hours ago), one of which caused BS, ITMY, and ETM watchdogs to trip. I restored the damping just now.

I worked on preparing for the c1iscaux upgrade a bit today.

1. Attachment #1: This shows where the 120 GB solid-state hard-drive and the 2 RAM cards (2GB each) are installed.
   • I found that it required considerable application of force to get the RAM cards into their slots.
   • Note: the 4GB RAM is broken up into two separate physical cards, each 2GB. The labeling is a bit confusing, as each card suggests it is by itself 4GB.
2. OS install for c1iscaux:
   • I followed Jon's instructions (and added some of mine to the wiki page to hopefully make this process even less thinking-intensive).
   • To be able to use the IP address 192.168.113.83, removed "bscteststand" from chiara martian.hosts and rev.113.168.192.in-addr.arpa as the last mention I could find of this machine was from 2009 (and I'm pretty sure it isn't an active unit anymore). I then restarted the bind9 process. 
   • The hostname for this machine is currently "c1iscaux3" for testing purposes, I will change it once we do the actual install.
   • There was an error in the installation instructions to allow incoming ssh connections - it is openssh-server that is required, not openssh-client. This has now been fixed on the wiki page instructions.
3. Acromag static IP assignment:
   • Assigned 2 ADCs (XT1221), 5 DACs (XT1541) and 5 sinking BIO units (XT1111) static IP addresses (and labelled them for easy reference) using the windows laptop and the Acromag IP config utility.
   • I saw no reason not to use the 192.168.114.yyy scheme for the Acromag subnet on this machine, even though c1auxex and c1vac both have subnets with this addressing prefix. For reasons unknown to me, Jon opted to use 192.168.115.yyy for the c1susaux Acromag subnet.
4. Followed the excellent step-by-step to install EPICS, Modbus and Asyn.
   • This took a while, ~1 hour, dominated by the building of EPICS. The other two took only a couple of minutes each.
   • The same combination suggested on Jon's wiki, of Modbus R2-11, EPICS base-7.0.1 and asyn4-33, are the most current at the time of installation.
   • Couple of typos that prevented straight up copy-pasting were fixed on the wiki.
5. Playground for testing new database files:
   • made a directory /cvs/cds/caltech/target/c1iscaux3 and copied over the .db files from /cvs/cds/caltech/target/c1iscaux and /cvs/cds/caltech/target/c1iscaux2 over.
   • Johannes said he did not develop any code to automate the process of translating the old .db files into the new ones for the Acromag - I won't invest the time in developing any either as I think just manually editing the files will be faster. 
   • I think I will follow the c1susaux convention of grouping .db files by the physical electronics system where possible (e.g. REFL11 channels in one file, CM channels in one file etc), as I think this makes for easier debugging.
   • There is an old "PZT_AI.db" file which I think consists completely of obsolete channels.
6. Next steps:
   • Wire up the crate [Chub]
   • Make the database files and modbus files for talking to the Acromags on the internal subnet [Gautam], check the .db files [Koji]
   • Wiring of whitening switching from P1 to P2 connector, Issue #1 in this elog (this will also requrie the installation of the DIN shrouds) [Koji]
   • Soldering of P2 interface boards [Gautam]
   • Bench testing [Gautam, Koji, Chub]
   • Installation and in-situ testing [Gautam, Koji, Chub]

All the required additional parts should be here by the end of the week - I'd like to aim for Wednesday 7/24 for the installation in 1Y3 and in-situ testing. While talking to Rana, I realized that we should also factor in the c1aux slow channels into this acromag crate - there is no need for a separate machine to handle the shutters and illuminators. But let's not worry about that for now, those channels can simply be added later.

[Kruthi, Gautam, Rana]

Gautam installed Atom text editor on Pianosa yesterday.

MC spot position measurement scripts (these can be found in /scripts/ASS/MC directory): 

• Changed the power threshold for MC2 lock loss check from 15000 to 12000 (volts) in the MeasureSpotPositions.py script. This is because, the C1:I00-MC_TRANS_SUM reads a value, usually, greater than 14000 and with 15000 as the threshold, the script will always say the MC isn't locked even though it is!. Also, to account for additional variation we have a margin of 2000.
• Issues with datetime: though MeasureSpotPositions.py was creating a .dat file, MC_spotMeasurement_history.py threw an error because the .dat file's name was not in the required format. I fixed this bug.
• Just running the MeasureSpotPositions.py doesn't enter the results into the log file, instead ./mcassMCdecenter should be run
• MC_spotMeasurement_history.py just plots the spot positions (in mm) vs days since 2013, using the log file. It still has some bugs

I've taken the MC2 analog camera down and put another GigE (unit 151) in its place. This is just temporary and I'll put the analog camera back once I finish the MC2 loss map calibration. I'm using a 25mm focal length camera lens with it and it gives a view of MC2 similar to the analog camera one. But I don't think it is completely focused yet (pictures attached).

...more to follow

gautam - Attachment #3 is my (sad) attempt at finding some point scatterers - Kruthi is going to play around with photUtils to figure out the average size of some point scatterers.

[koji, gautam]

Koji pointed out an important subtlety pertaining to the "LATCH ENABLE" signal line on the CM board. The purpose of this line is to smoothly facilitate the transition of a change in the "multi-bit-binary-outputs", a.k.a. "mbbo", that are controlled by MEDM gain sliders, to the analog electronics on the CM board. Why is this necessary? Imagine changing the gain from 7dB (=0111 in mbbo representation) to 8dB (=1000 in mbbo representation). In order to realize this change, all 4 bits have to change their state. But this almost certainly doesn't happen synchronously, because our EPICS interface isn't synchronous. So at some intermediate times, the mbbo representation could be 0100 (=4dB), or 1111 (=15dB), or many other possible values, which are all significantly different from either the initial value or the desired final state. This is clearly undesirable.

In order to protect against this kind of error, a Latched output part, 74ALS573, is used to buffer the physical digital logic levels from the switches in the analog gain stages. So in the default state, the "LATCH ENABLE" signal line is held "LOW". When a change happens in the EPICS value corresponding to a gain slider, the "LATCH ENABLE" state is quickly toggled to "HIGH", so as to enable the appropriate analog gain stages to be switched, and then again to "LOW", at which point the latch holds its output state. This logic is currently implemented by a piece of code called "latch.o", which is the compiled version of "latch.st", which may be found in /cvs/cds/caltech/target/c1iool0 where it presumably was written for the IMC servo board, but not in /cvs/cds/caltech/target/c1iool0  , which is where the CM board database files reside. The only elog reference I can find pertaining to this particular piece of code is from Alan, and doesn't say anything about the actual logic.

For the new c1iscaux, we need to implement this logic somehow. After discussion between Koji and me, we feel that a piece of python code is sufficient. This would continuously run in the background on the supermicro server machine. The channel hierarchy for each gain channes is as follows (I've taken the example of C1:LSC-CM_REFL1_GAIN):

• C1:LSC-CM_REFL1_GAIN ------ this is the channel tied to an MEDM slider, and so is a "soft" channel
• C1:LSC-CM_REFL1_SET ------- this is a "soft" channel that gets converted to an mbbo
• C1:LSC-CM_REFL1_BITS ------ this is a channel that actually controls (multiple) physical binary outputs on the Acromag

So the logic will be that it continuously scans the EPICS channel C1:LSC-CM_REFL1_GAIN  for a change in set value. When a change is detected, it has to update the C1:LSC-CM_REFL1_SET channel. In the next EPICS refresh cycle, this would result in the mbbo bits, C1:LSC-CM_REFL1_BITS , all changing to the appropriate values. After these changes have happened, we need to toggle the LATCH ENABLE in order to allow the changes to propagate to the analog gain stage switches. Need to think about what's the best way to do this.

I completed the translation of the .db files for the EPICS database records from the VME notation to the Acromag/Modbus/Asyn notation. The channels are now organized into 5 database files, located in /cvs/cds/caltech/target/c1iscaux3/,  for convenience:

1. C1_ISC-AUX_LSCPDs.db -------- This handles whitening gain, AA enable/bypass, Demodulator FE, and PD Interface Board channels for REFL11, REFL55, REFL33, REFL165, POP22, POP110, POX11, POY11, AS55 and AS110 photodiodes.
2. C1_ISC-AUX_CM.db -------------- This handles all channels for the CM board. The mbbo addressing notation needs to be checked.
3. C1_ISC-AUX_QPDs.db ----------- This handles all channels for the IPPOS QPD.
4. C1_ISC-AUX_ALS.db ------------- This handles all channels for the IR ALS DFD LO and RF power monitoring.
5. C1_ISC-AUX_SPARE.db ---------- This handles the unused channels for the various whitening, AA and PD interface boards.

For reasons unknown to me, the database files in the other Acromag system target directories (e.g. c1susaux, c1auxex) all had 755 level access permission - maybe this is required for systemctl to handle the EPICS serving? Anyways, I upgraded the permission level of the above 5 files using chmod.

There are almost certainly typos / other errors, and I may have missed copying over some soft/calibrated channels, but I hope that this way of grouping by subsystem will make the debugging less painful. Once Chub connects up the power lines to the Acromags, I will run the soft tests. For this purpose, I've also made a C1_ISC-AUX.cmd file and a C1_ISC-AUX.env file in the above target directory, and also made the modbusIOC.service file in /etc/systemd/system on the supermicro.

Now that the .db files were prepared, I wanted to test for errors. So I did the following:

1. Acromags were mounted on the DIN rails. Attachment #1 shows the grouping of ADC, DAC and BIO units. They are labelled with their IP addresses.
2. Wiring of power:
   • Chub had already prepared the backplane with the power connectors, switches and indicator LEDs.
   • So I just had to daisy chain the +24 V (RED) and GND (BLACK) terminals for all the acromags together, which I did using 24 AWG wire (we may want to use heavier gauge given the current draw).
3. Ethernet cables were used to daisy chain the network connectivity between the various units.  Attachment #1 shows the current state of the chassis box.
4. Front panel pieces were attached and labelled, see Attachment #2. 
   • I found it was sufficient to use the front - we may use the rear panel slots when we want to add connections for controlling the c1aux machine channels.
   • The D15 P2 connector panel for the CM board will arrive tomorrow and will be installed then.
5. Entire setup was connected to power and ethernet, see Attachment #3. 
   • As usual, the current draw is significant for the collection of Acromags, I got around this problem by using the bench supply to "Parallel" mode to enhance the current driving capacity.
   • For the ethernet connection, I used the office space port #6, which I connected at the network rack end to the eth1 port of the Supermicro.

All the Acromags are seen on the 192.168.114 subnet on c1iscaux3 - however, when I run the modbusIOC process, I see various errors in the logfile , so more debugging is required. Nevertheless, progress.

Update 2245: Turns out the errors were indeed due to a copy/paste error - I had changed the IP addresses for the ADCs from the .115 subnet c1susaux was using, but forgot to do so for the DACs and BIOs. Now, if I turn off the existing c1iscaux so that there aren't any EPICS clashes, the EPICS server initializes correctly. There are still some errors in the log file - these pertain to (i) the mbbo notation, which I have to figure out, and (ii) the fact that this version of EPICS, 7.0.1, does not support channel descriptions longer than 28 characters (we have several that exceed this threshold). I think the latter isn't a serious problem.

Getting closer... Note that I turned off the c1iscaux VME crate to prevent any EPICS server clashes. I will turn it back on tomorrow.

[Kruthi, Yehonathan, Gautam]

Today evening, Yehonathan and I aligned the MC2 cameras. As of now there are 2 GigEs in the MC2 enclosure. For the temporary GigE (which is the analog camera's place), we are using an ethernet cable connection from the Netgear switch in 1x6. The MC2 was misaligned and the autolocker wasn't able to lock the mode cleaner. So, Gautam disabled the autolocker and manually changed the settings; the autolocker was able to take over eventually.

Rana and I talked about some (genius) options for the large range DC bias actuation on the SOS, which do not require us to supply high-voltage to the OSEMs from outside the vacuum.

What we came up with (these are pretty vague ideas at the moment):

1. Some kind of thermal actuation.
2. Some kind of electrical actuation where we supply normal (+/- 10 V) from outside the vacuum, and some mechanism inside the chamber integrates (and hence also low-pass filters) the applied voltage to provide a large DC force without injecting a ton of sensor noise.
3. Use the blue piers as a DC actuator to correct for the pitch imbalance --- Kruthi and Milind are going to do some experiments to investigate this possibility later today.

For the thermal option, I remembered that (exactly a year ago to the day!) when we were doing cavity mode scans, once the heaters were turned on, I needed to apply significant correction to the DC bias voltage to bring the cavity alignment back to normal. The mechanism of this wasn't exactly clear to me - furthermore, we don't have a FLIRcam picture of where the heater radiation patter was centered prior to my re-centering of it on the optic earlier this year, so we don't know what exactly we were heating. Nevertheless, I decided to look at the trend data from that night's work - see Attachment #1. This is a minute trend of some ETMY channels from 0000 UTC on 18 July 2018, for 24 hours. Some remarks:

1. We did multiple trials that night, both with the elliptical reflector and the cylindrical setup that Annalisa and Terra implemented. I think the most relevant part of this data is starting at 1500 UTC (i.e. ~8am PDT, which is around when we closed shop and went home). So that's when the heaters were turned off, and the subsequent drift of PIT/YAW are, I claim, due to whatever thermal transients were at play.
2. Just prior to that time, we were running the heater at close to its maximum rated current - so this relaxation is indicative of the range we can get out of this method of actuation.
3. I had wrongly claimed in my discussion with Rana this morning that the change in alignment was mostly in pitch - in fact, the data suggests the change is almost equal in the two DoFs. Oplev and OSEMs report different changes though, by almost a factor of 2....
4. The timescale of the relaxation is ~20 minutes - what part(s) of the suspension take this timescale to heat up/cool down? Unlikely to be the wire/any metal parts because the thermal conductivity is high? 
5. In the optimistic scenario, let's say we get 100 urad of actuation range - over 40m, this corresponds to a beam spot motion of ~8mm, which isn't a whole lot. Since the mechanism of what is causing this misalignment is unclear, we may end up with significantly less actuation range as well.
6. I will repeat the test (i.e. drive the heater and look for drift in the suspension alignment using OSEMs/Oplev) in the afternoon - now I claim the radation pattern is better centered on the optic so maybe we will have a better understanding of what mechanisms are at play.

Also see this elog by Terra.

Attachment #2 shows the results from today's heating. I did 4 steps, which are obvious in the data - I=0.6A, I=0.76A, I=0.9A, and I=1.05A.

In science, one usually tries to implement some kind of interpretation. so as to translate the natural world into meaning.

[chub, gautam]

Bulb replaced. Projector is back on.

The control room UPS started making a beeping noise saying batteries need replacement. I hit the "Test" button and the beeping went away. According to the label on it, the batteries were last repalced in March 2016, so maybe it is time for a replacement, @Chub, please look into this.

I did a whole lot of hyperparameter tuning for convolutional networks (without 3d convolution). Of the results I obtained, I am attaching the best results below.

Define "best"?

The lower the power of the error signal (difference between the true and predicted X and Y positions), essentially mse, on the test data, the better the performance of the model. Of the trained models I had, I chose the one with the lowest mse.

Attached results:

1. Attachment 1: Training configuration
2. Attachment 2: Predicted motion along the Y direction for the test data
3. Attachment 3: Predicted motion along the Y direction for the training data
4. Attachment 4: Learning curves
5. Attachment 5: Error in test predictions
6. Attachment 6: Video of image histogram plots
7. Attachment 7: Plot of percentage of pixels with intensity over 240 with time

(Note: Attachment 6 and 7 present information regarding a fraction of the data. However, the behaviour remains the same for the rest of the data.)

Observations and analysis:

1. Data:
   1. From attachemtns 2, 3, 5: Maximum deviation from true labels at the peaks of applied dither/motion. Possible reasons:
      1. Stupid Cropping? I checked (by watching the video of cropped frames, i.e visually) to ensure that the entire motion of the beam spot is captured. Therefore, this is not the case.
      2. Intensity variation: The intensity (brightness?) of the beam spot varies (decreases) significantly at the maximum displacement. This, I think, is creating a skewed dataset with very few frames with low intensity pixels. Therefore, I think it makes sense to even this out and get more data points (frames) with similar (lower) pixel intensities. I can think of two ways of doing this:
         1. Collect more data with lower amplitude of sinusoidal dither. I used an amplitude of 80 cts to dither the optic. Perhaps something like 40 is more feasible. This will ensure the dataset isn't too skewed.
         2. Increase exposure time. I used an exposure time of 500us to capture data. Perhaps a higher exposure time will ensure that the image of the beam spot doesn't fade out at the peak of motion.
   2. From attachment 5, Saturated images?: We would like to gun for a maximum deviation of 10% (0.1 in this case) from the true values in the predicted labels (Tbh, I'm not sure why this is a good baseline, I ought to give that some thought. I think the maximum deviation of the OpenCV thing I did at the start might also be a good baseline?). Clearly, we're not meeting that. One possible reason is that the video might be saturated- (too many pixels at 255, bleeding into surrounding pixles) leading to loss of information. I set the exposure time to 500us precisely to avoid this. However, I also created videos of the image histograms of the frames to make sure the frames weren't saturated (Is there some better standard way of doing it?). From attachements 6 and 7, I think it's evident that saturation is not an issue. Consequently, I think increasing the exposure time and collecting data is a good idea.
2. The network:
   1. From attachment 4: Training post 25 epochs seems to produce overfitting, though it doesn't seem too terrible (from attachments 2 and 3). The network is still learning after 75 epochs, so I'll tinker with the learning rate, dropout and maybe put in annealing.
   2. I don't think there is a need to change the architecture yet. The model seems to generalize okay (valdiation error is close to training error), therefore I think it'll be a good idea to increase dropout for the fully connected layers and train for longer/ with a higher learning rate.

P.S. I will also try the 2D convolution followed by the 1D convolution thing now. 

P.P.S. Gabriele suggested that I try average pooling instead of max pooling as this is a regression task. I'll give that a shot.

For some reason, rossa's Xdisplay won't start up anymore. This happened right after the UPS reset. Koji and I tried ~1.5 hours of debugging, got nowhere.

Summary:

The database files for C1ISCAUX seem to work file - the exception being the mbbo channels for the CM board.

Details:

This was just a software test - the actual functionality of the channels will have to be tested once the Acromag crate has been installed in the rack. One change I had to make on the MEDM screen for the LSC PD whitening gains was to get rid of the "NMS" suffix on the EPICS channel names for whitening gain sliders/drop-down-menus. I suspect this has to do with the EPICS version we are using, 7.0.1. Furthermore, AS165 and POP55 no longer exist - I hold off removing them from the MEDM screen for the moment.

Next steps:

From the software point of view, the major steps are:

1. Fix the mbbo channel notation in the database files
2. Write and test the latch enabling code
3. Figure out what scripted tests can be done to test the functionality of the new Acromag box.

I am stopping the EPICS server on the new machine and restarting the old VME crate over the weekend.

I'm not able to get trends of the TM adjustment test that Rana had asked us to perform, from the dataviewer. It's throwing the following error:

Connecting to NDS Server fb (TCP port 8088)
Connecting.... done
Server error 7: connect() failed
datasrv: DataWrite failed: daq_send: Resource temporarily unavailable
T0=19-07-20-01-27-39; Length=600 (s)
No data output.

What channels are you trying to read?

Summary:

SnapPy scripts made to work on Pianosa.

Details:

Of course rossa was the only machine in the lab that could run the python scripts to interface with the GigE camera. And it is totally bricked now. Lame.

So I installed several packages. The key was to install pypylon - if you go to the basler webpage, pypylon1.4.0 does not offer python2.7 support for x86_64 architecture, so I installed pypylon1.3.0. Here are the relevant lines from the changelog:

Camera server is running on a tmux session on pianosa. But it keeps throwing up some gstreamer warnings/errors, and periodically (~every 20 mins) crashes. Kruthi tells me that this behavior was seen on Rossa as well, so whatever the problem is, doesn't seem to be because I missed out on installing some packages on pianosa. Moreover, if the server is in fact running, I am able to take a snapshot - but the camera client does not run.

1. Make the MSE a subplot on the same axes as the time series for easier interpretation.
2. Describe the training dataset - what is the pk-to-pk amplitude of the beam spot motion you are using for training in physical units? What was the frequency of the dither applied? Is this using a zoomed-in view of the spot or a zoomed out one with the OSEMs in it? If the excursion is large, and you are moving the spot by dithering MC2, the WFS servos may not have time to adjust the cavity alignment to the nominal maximum value.
3. What is the minimum detectable motion given the CCD resolution?
4. Please upload a cartoon of the network architecture for easier visualization. What is the algorithm we are using? Is the approach the same as using the bright point scatterers to signal the beam spot motion that Gabriele demonstrated successfully?
5. What is the significance of Attachment #6? I think the x-axis of that plot should also be log-scaled.
6. Is the performance of the network still good if you feed it a time-shuffled test dataset? i.e. you have (pictures,Xcoord,Ycoord) tuples, which don't necessarily have to be given to the network in a time-ordered sequence in order to predict the beam spot position (unless the network is somehow using the past beam position to predict the new beam position).
7. Is the time-sync problem Koji raised limiting this approach?

<Adding details>

See Attachment #2.

Training dataset:

1. Peak to peak amplitue in physical units: ?
2. Dither frequency: 0.2 Hz
3. Video data: zoomed in video of the beam spot obtained from GigE camera 198.162.113.153 at 500us exposure time. Each frame has a resolution of 640 x 480 which I have cropped to 350 x 350. Attachment #1 is one such frame.
4. Yes, therefore I am going to obtain video at lower amplitudes. I think that should help me avoid the problem of not-nominal-maximum value?
5. Other details of the training dataset:
   1. Dataset created from four vides of duration ~ 30, 60, 60, 60 s at 25 FPS.
   2. 4032 training data points
      1. Input (one example/ data point): 10 successive frames stacked to form a 3D volume of shape 350 x 350 x 10
      2. Output (2 dimensional vector): QPD readings (C1:IOO-MC_TRANS_PIT_ERR, C1:IOO-MC_TRANS_YAW_ERR)
   3. Pre-processing: none
   4. Shuffling: Dataset was shuffled before every epoch
   5. No thresholding: Binary images are gonna be of little use if the expectation is that the network will learn to interpret intensity variations of pixels.

Do I need to provide any more details here?

?

see attachment #4.

I wrote what I think is a handy script to observe if the frames are saturated. I thought this might be handy for if/when I collect data with higher exposure times. I assumed there was no saturation in the images because I'd set the exposure value to something low. I thought it'd be useful to just verify that. Attachment #3 has log scale on the x axis.

I'm running the MC2 loss map scripts on pianosa now. The camera server is throwing an error and is not grabbing snapshots :(

Update: I finished taking the readings for MC2 loss map. I couldn't get the snapshots with the script, so I manually took some 4-5 pictures.

Can you please be more specific about what the error is? Is this the usual instability with the camera server code? Or was it something new?