Archive for June, 2017

Week 7 at an Unbootcamp

Wednesday, June 21st, 2017

First week under the new model

As reported last week, Learners Guild in Oakland, California, has just made a major change in its software-development curriculum. This week was the first week of the new model’s implementation.

Under the new model, coding tests and technical interviews allow Learners to progress through 5 “phases”. This is the transition week. All 120 or so Learners are undergoing coding tests and interviews simultaneously. We were all tested from Monday morning until noon today (2.5 days), and then we are each getting a half-hour teleconferenced technical interview sometime during the rest of this week, followed by a decision on our phase placement. It is impressive that the Guild can handle this process for all Learners in the same week.

My own interview took place this afternoon. I was my interviewer’s first interviewee. I survived with my self-respect intact, unlike job interviewees in software who are often reportedly humiliated and demeaned by their interviewers (see, e.g., William Poundstone’s How Would You Move Mount Fuji?.) The interview was friendly and relaxed. As expected, I was mainly asked to explain how and I had solved problems and why I had done so as I did. Occasionally I was also asked what, if anything, I knew about related other methods. If I imagine how I would want to interview a person in the same situation, to help my organization decide where best to place the person in the curriculum, that’s how it went.

Before the interview I spent 2.5 days performing 3 assigned tasks that I doubted I could finish. Somehow, without losing much sleep, I did manage to complete them all. To do that, I had to look up a lot of new information and reawaken some dormant know-how.

I also had to make a decision to violate one of the methodological requirements. It said: Do X, using tool Y. Well, I tried, but after several hours decided I would not have time to figure tool Y (a new tool for me) out. I decided to do X with method Z instead. It’s like, I rationalized, deciding to serve baked potato rather than broiled ham to a customer who has asked for broiled potato.

I was tested and interviewed for admission into phase 3. Phase 3 out of 5? After only 6 weeks out of 40? Yes, because I took seriously, and trusted, the staff’s recommendation to choose a phase ambitiously. I appear to have been more trusting in this respect that my fellow junior Learners, who appear uniformly to have chosen to be interviewed for phase 2. Not terribly consequential, since, if I don’t get admitted into phase 3, I’ll be in phase 2, just like those who interviewed for phase 2, and we’ll all have 8 weeks to get into phase 3. Now that it’s done, I don’t regret trusting the advice to be ambitious. This test kept me doing research, testing, and debugging, and I wound up, probably, learning more from the experience than if I had been working on the prior test’s problems. If it turns out that I almost got into phase 3, then I’ll have the option to re-interview for it as soon as I feel ready.

The extent of the curricular revolution at the Guild became clearer today when the new curriculum’s outline was disclosed to the Learners. It’s a truly magnum opus, although still in progress. A plethora of interfaces have been consolidated into a portal, where we can now see all the phases, the skills they develop, the project modules that we can work on, which modules are good for which skills, an outline of topic lectures to be offered, a photo gallery of all Learners and staff members, a link from each gallery entry to methods for contacting the person, and the basic operation manual for life at the Guild (a.k.a. the Guide). As of this writing, there are 208 skills listed for phase 1, and 160 skills listed for phase 2. Click on a skill, and you go to a page showing which modules you can learn or exercise that skill with. Click on one of those module titles, and you get a page on that module, containing the skills it develops (classified into categories), a glossary of related terms, terms useful in searches for related help, links to many related exercises on the web, and links to many related resources. What you don’t find on the page is a set of instructions giving you a programming problem to solve. That’s the one thing you might have expected in a traditional teaching/learning institution. Its absence is a signal as to how the Guild differs. You’ll get programming problems when you apply to move into the next phase, but in the meantime you are doing your own learning, with published support from the curriculum, personal support from professional engineers, and about 120 other Learners you can always ask for help from.

Now that my interview is completed, I have Thursday and Friday to choose some topics to get familiar with. The new curriculum mentions about 400 such topics, and a few of them became salient during my test and interview. I won’t run out of topics any time soon.


Week 6 at an Unbootcamp

Friday, June 16th, 2017

No ordinary week

“Cycle 47” was a special week in my career makeover at Learners Guild. The old Guild is no more; enter the new Guild.


First, the week was abbreviated, for me, by a day and a half of jury duty. I was confident I would never survive the peremptory challenges in any jury selection: I’m too iconoclastic and steeped in litigation. This was a residential eviction case, where the defendant was claiming retaliation. I myself had just moved out of a co-op in anticipation that its members would perpetrate retaliatory eviction against me. How could the landlord’s attorney let me onto that jury? But she did. The tenant’s attorney asked whether, despite my involvement in numerous prior eviction cases, I could be unbiased in this case, and I said no, I could not be unbiased. But he still didn’t challenge me! Both attorneys exhausted their challenge allotments on persons who, in my lay judgment, were less risky for their clients, but not on me. So I actually got onto the jury. We heard our instructions and witnessed a couple of hours of botched testimony by a landlord witness, who said his firm normally complies with the law on evictions but admitted he didn’t know whether it had complied in this case. Why this case ever reached trial was a mystery to me. And the next morning we were informed that the case had settled, so we were let loose. Before leaving the courtroom, I asked the landlord’s attorney whether her client couldn’t have prepared better for this trial, and she agreed.

Back at Learners Guild, I resumed work on my Goal on Wednesday. “My”, because the Algorithm had assigned me to my second-choice (solo) Goal rather than my first-choice (team) Goal. Had I been on a team, I would have asked to have my jury duty rescheduled, in fairness to my teammate.

The nitty gritty

The Goal required solving 8 logical problems, called “coding exercises”:

  1. Scrabble tally: Given a list of tiles already used, return a tally of all letters, grouped by the counts of their remaining tiles. I chose to implement it generically, so it would work for any language’s version of Scrabble, or any other structurally identical situation.
  2. Word lookup in a string: Given a string (i.e. a line of text) and a number, return the word that is at the number’s position in the string, where the first position is zero. I chose to implement it with the XRegExp library, so it would work for any language, not just English. Whether for English or generically, one must define “word”. For this exercise, I defined the words of a string as what are left when we treat any sequence of characters matching the regular expression “[^\pL\pN]*\pZ[^\pL\pN]*” as a delimiter. That means any sequence of 0 or more characters other than letters and/or numbers, provided that it includes at least 1 separator character (using Unicode’s definitions of letters, numbers, and separators). There is no universally satisfactory rule defining “word” in all languages, or even in English. Mine was a quick-and-dirty rule. It classifies “M&M.” as a word in “He ate an M&M.”, and “Mr” as a word in “Mr. Smith”.
  3. Translator between (ordinary) decimal numbers and binary numbers with the Fibonacci sequence as the base. (We who love to solve problems don’t ask, “What am I ever going to need this for?”.)
  4. Function that removes vowels, defined as “a”, “e”, “i”, “o”, “u”, and all non-a-to-z characters (even A-to-Z), from a string. Defined naïvely and Anglocentrically, so I couldn’t try to universalize it. Lucky for me; that could have taken a lifetime.
  5. A search tool to find a value inside a JSON string. JSON seems to have taken over supremacy from XML as a format for exchanging complex bodies of information. Its advocates say it’s “much simpler”.
  6. Date reformatter: Converts from any of 6 formats to the ISO 8601 standard format, like “2017-06-16”.
  7. Music player: A web page that, as soon as you open it, plays the notes of an octave and, optionally, a chord. Why was this included here? Sure, it’s coding, but it also involves researching how to generate and play sound. That’s what took the most time for me.
  8. Parenthetical redundancy eliminator: Given a string, return the same string, with any redundant pairs of parentheses removed. Not as simple as it may seem. For example, “ak⁾vo (((water [[(H₂O)]])))” contains no redundant pairs of parentheses, because brackets and full-width parentheses are not (in my solution) classified as parentheses.

That, then, is a taste of the kind of thinking we learn to practice at the Guild.

Would the detour hurt my grade?

Given the interruption, I doubted I could complete the 1-week Goal’s requirements by Friday afternoon. My coach reassured me that my completion score on this Goal would not suffer from my jury-duty absence, because scores were adjusted for time off. In any case, I managed to finish the 8 problems on Thursday, so I used the remaining time to start some of the Goal’s optional work.

Anyway, grades didn’t exist at the Guild. Statistics and levels did, though, and it’s hard to avoid assimilating any quantitative indicator into one’s mental model of grading at school. Even without grades, I have been seeing a considerable amount of grade anxiety among Learners, attached to what are the closest approximations to grades.

Revolution from above

Little did my coach or I know that the week’s work was not going to be scored for completion at all. We did get a hint of this on Monday, when the staff advised us not to worry about completion scores that week and to focus only on learning as much as we could. But the deeper reason for this advice was revealed only on Friday.

Learners Guild solicits comments from its Learners, and it gets many unsolicited comments as well. Its staff makes changes in the process in response to those comments.

On Friday afternoon, when I normally would have been composing this blog entry (well, a blog entry, since the content would have differed), we spent 3 hours learning of, asking questions about, and discussing a new batch of changes to the system. This one was no set of incremental adjustments. It was fundamental.

The Guild has now jettisoned its entire apparatus of statistical progress (and regress) indicators, which had been refined for months and was about to be re-refined. By one staffer’s own testimony, the repeated attempts to make that system satisfactory had led to a conviction that doing so was hopeless, so, in mid-tweak, the system was trashed.

The set of “levels” through which Learners moved on the basis of their statistical indicators has likewise been thrown overboard. This vertical (higher and lower) metaphor has been replaced by a horizontal one: “phases”.

Phases might seem like a mere renaming of levels, but they are not:

  • The phases differ in their content. Phases 1 and 2 are self-paced individual work devoted to mastery of prescribed fundamentals. In those phases, no more do Learners vote on preferred projects; instead, there is a fixed set of skills to be acquired. Phase 3 is team-based small projects with choice, like what I have been mostly doing up to now. Phase 4 is collaborative work on real (not just realistic) open-source projects. Those two project phases, 3 and 4, are not rigidly tied to a 1-week cycle. And Phase 5 is preparation for the reality of the external market for the Guild’s product, web application developers.
  • Admission into the next phase depends on a 1-week testing and interviewing filter. We should expect on average to be ready for that after about 5 weeks in a phase, but we must get into the next phase in no later than 8 weeks; otherwise, our progress is not satisfactory. If we don’t get admitted on our first attempt, we get an additional opportunity to try. Phase 5 is not subject to that limit if it would eject us before our 40 weeks at the Guild has come to an end.
  • Learner-to-learner coaching is no longer a formal element of the system. Instead, we are supported by “Software Engineering Practitioners”, who are Guild employees. To see the qualifications of an SEP, look at the current hiring announcement. If you yourself are qualified, please apply and mention my blog as your referral source. While coaches are no more, the Guild hopes that its abandonment of microstatistical indicators will make Learners more relaxed about taking time from their work to help other Learners. Thus, if it works, coaching will become culturally pervasive, instead of being an assigned temporary role.

The path from here to there

The Guild seems to have defined a reasonably smooth transition for existing Learners into the new order. This week we are on a pre-scheduled 1-week break. Next week we all undergo a “sorting” process. To prepare, each of us has notified the Guild which phase he or she asks to be evaluated for. Starting Monday morning, we work individually on projects constituting take-home tests for the phases we have chosen. Starting Wednesday afternoon, we get interviewed by SEPs on the solutions we produced, to evaluate our (in)ability to explain and reason about our own solutions and possible alternatives. At the end of the week we learn whether we have been admitted to the phases we selected; if not, we are admitted to the preceding phases.

The Guild program lasts, nominally, 40 weeks (plus 1 or 2 break weeks). If the transition places a Learner into a phase that makes the program longer, then it is extended for that Learner.

I have been at the Guild for 6 weeks now. That should qualify me to get into Phase 2. But the Guild advises us to try to get into the most advanced phase we “think” we “can” get admitted to. I have probed for clarity about that, and I believe the intended meaning is the most advanced phase we believe the SEPs have a nontrivial probability of letting us into. So, if my subjective probability of success in getting into Phase 3 is 30%, I should try for Phase 3. That’s close to my subjective reality, so I have asked to be tested for Phase 3. If I were to succeed, I would feel a need to backfill my fundamentals, and I’d be doing so under pressure to complete current projects, but that’s probably typical of the real software-development world.

Who’s the guinea pig?

Abrupt change can be exhilarating or frightening, and the changes can benefit some while harming others. The Learner population is diverse in many ways, so the personal impacts of this redesign will vary.

At least one Learner has said that the Guild feels like a laboratory in which she is one of the subjects in an experiment.

There is a basis for that impression. The Guild exists only in Oakland but aspires to replicate itself. We are testing each iteration of its design and helping the Guild decide what to keep and what to change. The system must work for us before it can be scaled up and out.

While acknowledging the validity and empirical basis for the sense that the Guild is experimenting on us, I have a different feeling. For me, yes, there is an experiment going on, but in that experiment I am the experimenter and the Guild is the subject.

I am conducting an experiment (or, perhaps better, pilot study) testing the hypothesis that there is no chronological limit to professional rejuvenation, even in information technology. I invited the Guild to join that experiment, and it agreed. As time passes, I keep providing stimuli to the Guild, and it keeps responding. Step by step, I’m discovering what makes it tick and how to get my needs satisfied by this subject, who is simultaneously a subject in 120 other engineers’ experiments, testing similar but not identical hypotheses. In this interaction, I perceive more autonomy in myself than in the Guild.

I have also heard the suggestion that the Guild may be breaking its contract with Learners when it changes its system. I have my doubts about that, and in fact I would probably entertain such a notion if the Guild refused to change its system, because, arguably, that would be negligence.

But these are subjective matters. If you are considering applying to the Guild, considering the testimony of other Learners and me may help you predict what your own interpretations would be. If you are risk- and change-acceptant, I believe that’s a good sign of compatibility. Perhaps only such persons are a good fit for contemporary software development work anyway.


Week 5 at an Unbootcamp

Friday, June 2nd, 2017

“Cycle 46” is ending today at Learners Guild. Each cycle lasts a week. For me, this is week 5. Cycle 46 is a 4-day week, because of Memorial Day, so there has been a bit of pessimism in the air about whether the Learners would have enough time to complete their various projects. But not much. The Guild reminds us not to over-focus on our completion statistics.

Week 5 was my first encounter here with rejection. The Algorithm gave me my first-choice Goal to work on, together with another Learner, but he had not voted for it or, in fact, for any 2-person Goal. He wanted—desperately—to work on a 1-person Goal and asked me to give up my assignment so he could get a Goal that he wanted. It was, he said, part of a self-designed curriculum. I didn’t have the authority to change my assignment, but the staff does. I agreed to go along with his plan. The staff probed him for justification, but gently, and he didn’t yield, so the staff reassigned us—me to a 1-person Goal that was on my agenda. An agreeable outcome from my perspective: I’ve been spared during this 4-day week from negotiating roles with a teammate, and particularly with a reluctant and perhaps resentful one.

Is it beneficial for the Algorithm to impose its unknown rules, including some inevitable unpredictability, as it accepts or rejects the Goals that Learners voted for? Welcome to the real world. I have to assume that I, too, shall find myself some weeks working on Goals I haven’t voted for. So be it. We’re learning to do work that the world wants done (and is willing to pay for). The Algorithm is my boss, and bosses can be tempermental. We need practice in accommodating that.

I managed to complete the specifications on Thursday morning, so I had time to extend the project. Among the topics I explored further were execution efficiency and internationalization.

What’s most efficient isn’t obvious at first glance. Suppose you want the computer to inspect a sequence of characters (i.e. a “string”) and report whether it satisfies some criterion. The criterion posed in this case was whether the string is a palindrome: a string that is identical to its inversion. For example, “pool” is not a palindrome, but “poop” is. How can you instruct the computer to check this efficiently? One answer is: Tell it to make a inverted copy of the string and then to report whether the two strings are identical. It turns out that this is not the most efficient thing to do, at least in JavaScript.

  • If you think about it, you can see that this method requires the computer to do twice as much work as necessary. A comparison of “pinonip” with its inversion, i.e. “pinonip”, compares “o” with “o” once, but compares “p” with “p”, “i” with “i”, and “n” with “n” twice each. But checking the first 3 characters is enough: if “pin” is the same as “pin”, then the full string must be a palindrome. Any non-palindrome, such as “pinonap” (inversion “panonip”) will be detected within the first 3 comparisons.
  • Once you decide to compare only the first half with the second half’s inversion, how do you do that? Inverting the second half and then checking whether the two half-strings are identical is one method. But it turns out to be inefficient. It requires creating a new string. Another option is to compare the existing half-strings character by character, counting forward in one and backward in the other, using a “loop”. That runs faster. On my computer, it runs about twice as fast. To answer this question empirically, I needed to run my program on a string about 10 million characters long; much shorter strings were processed so fast that it was hard to measure the difference.

As for internationalization, the Goals I have seen so far are agnostic about the multiplicity of scripts, and they seem to reflect a tacit assumption that text is always in the Latin script and limited to the printable subset of the ASCII character set (roughly what you see embossed on the keys of a U.S. keyboard). The palindrome problem’s specifications told us to remove spaces and punctuation marks from the string before checking whether it is a palindrome. It also told us to convert all upper-case characters to the corresponding lower-case ones. Its sole example was limited to ASCII characters. I made my solution international by defining “spaces and punctuation” as all characters other than those classified by the Unicode Standard as “letters” and “numbers”. To make JavaScript implement those categories, I installed a package named XRegExp in my project. Some programming languages do that without extra packages, but JavaScript, in its current version, still needs that external tool.

The help I received from more expert personnel here included answers to my specific questions when I failed to find them on my own, but also suggestions that went beyond answers. A suggestion like “I think your solution is not as efficient as one that loops through the existing strings” was enough to send me on my way investigating the issue. Formal instruction this isn’t. My hunch is that it is more effective. For one thing, I never doze off during it.