Speech:Spring 2015 Benjamin Leith Log


 * Home
 * Semesters
 * Spring 2015
 * Proposal
 * Report
 * Information - General Project Information
 * Experiments - List of speech experiments

Week Ending February 3, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 10, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 17, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 24, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 3, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 10, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 24, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 31, 2015
Wednesday, March 25th

Met with group following spring break. A number of things were discussed. One of the foremost parts of this discussion included a discussion of the ways we can share "confidential" group information in a way that the other group could not see. The only concrete decision made on this was to use codewords (or code-phrases) to refer to our specific sub-tasks in a way that wont tip off the other group.
 * Overview:

In addition, we broke into teams of two, in what Sam referred to as a "Buddy System". This divided the group into teams of 2, who each individually are responsible jointly for identifying and executing/testing a variable change or other experiment hypothesis each week. This system will enable the group to, in theory, collectively make 5 or more individual changes per week, for a total of 25 "hypotheses" (or broadly speaking, anywhere from 25 to 50 experiments total - a pretty hefty number).

-Proceed with group-approved (or, Sam-approved) hypothesis testing in conjunction with "buddy-system" team-member
 * Task:

-Established system for reliable communication with group-mates (myself included)
 * Results:

-Devise and execute Test/Train with the criteria discussed during the meeting. At the time of the meeting there were two possible changes (we'll call them T1 and T2, see group log), one of which has since been ruled out on technical grounds.
 * Plan:

None yet, the meeting went actually better and more productively than expected.
 * Concerns:

I have a feeling that "T2" could be a big deal. More on that to come.

Sunday, March 29th

Since the last entry, a few things have been shot down and some other items have been brought up for discussion.
 * Overview:

1) "T1" (density value modifications, "declassified" now) is not going to work the way we'd originally thought, for technical-limitation reasons.

2) "T2" above seems like it may be a valid lead. Its hard to give estimates on possible improvement yet, more on that below.

2) New proposed item, "T3", handled by Sam's sub-group, which should also be promising. I'm not directly involved here, but I was around for some of the research part initially.

-Proceed with group-approved (or, Sam-approved) hypothesis testing in conjunction with "buddy-system" team-member (still ongoing) -To fill in the group master log with a list of possible changes and cross them out as they get tried, or shot down at the research stage. If the group likes that approach, it may become our standard moving forward.
 * Task:

-While I'm upset that the density changes planned for "T1" didn't pan out, the possible "T2" changes look promising. Especially if we're getting error numbers in the low thousands at creation time as originally mentioned by Sam.
 * Results:

-Execute 125h train with "baseline" density value, and watch for possible indicators of "T2" issue. If any errors are reported here, and if these errors scale up with train length as we think they might, we could be on to something. If we can't squeeze 125h and a 250h train into the next few days before Wednesday, Stephen and I will present whatever we do have.
 * Plan:

Planning to train with appropriate "T2" changes in the coming days. It will need to be soon enough to have something meaningful to show at the meeting. At least the 125h train needs to be done by Wednesday, and thusly documented.
 * Concerns:

Monday, March 30th

Had some trouble getting onto Caesar and running a train today when I attempted, others experienced issues as well. I tried to walk group members through the process of setting up the VPN client, and I think this helped some of them. I'm still unable to authenticate personally, and I believe Sam (or maybe Stephen?) has an email into the systems group on this topic.
 * Overview:

On hold for technical difficulty, will investigate tomorrow when I'm on-site at UNH.
 * Task:

None yet.
 * Results:

Same as above.
 * Plan:

Maybe the Caesar move over break didn't go as smoothly as planned...?
 * Concerns:

Tuesday, March 31st

Caesar is back online and seemingly fully functional. I'm not 100% sure exactly what changes happened in the background, but all initial group members who reported issues can now access the server both remotely and locally.
 * Overview:

Need to regroup with Stephen to find out status. We have not yet been able to test our proposed changes in light of all the login difficulties and script execution problems. I read his log and it doesn't look like he got much further than I did.
 * Task:

None yet. Will regroup tomorrow.
 * Results:

Continue T2 plan for lowering baseline.
 * Plan:

Perhaps a bit of a slow start, but I'm not overly concerned. If Sam's train finishes as expected, Stephen and I could have our 125hr and 250hr train results by Thursday or Friday Night (dependent on train time).
 * Concerns:

Week Ending April 7, 2015
April 1, 2015

To work on a 256hr train with a density value of 64. Some of Sam's initial trains had been failing and along with script errors and login issues this has prevented our running so far. Stephen will be continuing with his 128hr train.
 * Task:

So far good. More updates to follow.
 * Results:

Identical to task, will verify T2 changes. None yet. Will update this log with progress.
 * Plan:
 * Concerns:

April 6, 2015

-Read group logs, nothing new to report personally. Easter has nicely derailed my plans. Substantial update hopefully expected tomorrow.

Week Ending April 14, 2015

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 21, 2015
Received non-train work update from Kenneth. The plan is to create a list of parameters that were changed to show improvement, or that hindered us, or had no apparent effect. This will be useful for end-of-project documentation purposes. More on this to follow as I get into it.
 * Weds, April 15th

Started work on the non-train task. It's slow going, given that I need to piece together everyone's logs and the group mail chain to do it. Nothing to report yet.
 * Fri, April 17th

Checked logs. Some other non-train work is in progress at this time in addition to my own.
 * Saturday, April 18th

Week Ending April 28, 2015
Non-train documentation work has started to pick up some steam. Using a derivative of the data gathered by Chris, along with some graphs I've put together from that data, I've started to filter back through and figure out what specifically was changed in each experiment. For things that had noticeable effects on our results, this process is markedly easier than items where there was no change... for no-change issues, I will probably need to talk to the group on Wednesday.
 * Monday, April 28th

The graphs I'm working from look something like this (Example data, not real train data): http://i.imgur.com/hlUdDSP.png - the methodology here is that each "pivot point" on the graph represents a change made by the group for either a positive or negative effect. On occasions where two or more trains were attempted (really, scored) on the same day, the average is taken. The overall effect produces a good picture of our changes in error rate over time.
 * Tuesday, April 29th

As part of my task, I will need to take these "pivot points" and compile a list of them, for either the positive or negative direction, for both "short" and "long" trains. This methodology will work well, except that it wont account for changes which were attempted but had no effect. I will be polling the group for a list of these tomorrow. From there, I will create one document, that hopefully wraps it all together, with dates that should correspond to Chris's information.

Week Ending May 5, 2015
I made some changes to the experiment results spreadsheet to add projection lines and reformat it, also pulled out some outlying results to give a clear, consistent picture of our results. I would post it here, but I'm pretty sure it would be immediately espionage'd.
 * Wednesday, April 29th