on 09-Sep-2022 (Fri)

2.1. Model architecture

To forecast future customer behavior, our model is trained using individual sequences of past transaction events, i.e., chronological accounts of a customer’s lifetime. The example in Table 2 describes one such customer’s transaction history over seven consecutive discrete time periods. 4 This particular individual makes a transaction in the first week, followed by one week of inactivity, then transacting for two consecutive weeks, and so on; in weeks 3 and 4 they also received some form of a marketing appeal. The two calendar components – the month and week indicators – represent time-varying contextual information which is shared across the individuals within a given cohort. In addition, in this example, we include also an individual time-invariant covariate (gender) and a time-varying, individual-level covariate (marketing appeals). This particular customer history can then be represented as a sequence of vectors with five elements: the input variable plus the four covariates. Individual-level covariates are strictly optional – in our empirical study, the Base model is built without any such variables. Whenever individual covariates are included, we label the model Extended. Note that the model is completely agnostic about further extensions: all individual-level, cohort-level, time-varying, or time-invariant covariates are simply encoded as categorical input variables, and are handled equally by the model. This property makes our model extremely flexible in dealing with diverse customer behaviors observed across multiple contexts and platforms

At this point, people take different paths. Some give up, labeling all “self-help” books a waste of time. Others decide it’s just a problem of remembering everything they read, and invest in fancy memorization techniques. And many people become “infovores,” force-feeding themselves endless books, articles, and courses, in the hope that something will stick.

I want to suggest an alternative to all the approaches above: what you read is good and useful and very important, you’re just reading it at the wrong time.

You’re reading about time management techniques now, but they will only be useful two years from now, when you become a manager and have much greater demands on your time., etc....

The challenge of knowledge is not acquiring it. In our digital world, you can acquire almost any knowledge at almost any time.

The challenge is knowing which knowledge is worth acquiring. And then building a system to forward bits of it through time, to the future situation or problem or challenge where it is most applicable, and most needed.

This is the job of a “second brain” — an external, integrated digital repository for the things you learn and the resources from which they come. It is a storage and retrieval system, packaging bits of knowledge into discrete packets that can be forwarded to various points in time to be reviewed, utilized, or deleted.

The four top-level categories of P.A.R.A. — Projects, Areas, Resources, and Archives — are designed to facilitate this process of forwarding knowledge through time.

• By placing a note in a project folder, you are essentially scheduling it for review on the short time horizon of an individual project
• Notes in area folders are scheduled for less frequent review, whenever you evaluate that area of your work or life
• Notes in resource folders stand ready for review if and when you decide to take action on that topic
• And notes in archive folders are in “cold storage,” available if needed but not scheduled for review at any particular time

Note that we have re-created the tickler file, except instead of strict time-based horizons (daily, weekly, monthly, annually), they are scheduled contingently — if X happens, when Y arrives, if I want to do Z, etc.

But now we turn to a more fundamental question: how are these packets made? Once we capture something, how do we structure the note so that it’s easily discoverable and usable in the future? How do we make sure what we’re saving today adds value to future projects, even when we can’t predict or even imagine what those projects might be?

That is the job of Progressive Summarization.

There are two primary schools of thought on how to organize a note-taking program (or really any body of information, but I’ll use terms specific to note-taking apps):

Tagging-first approaches argue that there should be no explicit hierarchy of notes, notebooks, and stacks. Notes are envisioned as an ever-changing, virtual matrix of interconnected, free-floating ideas. Because many tags can be applied to one note, there are multiple pathways to discover any given note. Locating notes in specific notebooks and folders is seen as limiting and static.

Although tags have their uses, I don’t believe they work as a primary organizational system. In my experience, relying on tagging is too fragile and requires too much maintenance, spreading attention too uniformly across all notes whether or not they are truly valuable. The virtual matrix sounds cool and futuristic, but our minds are not made to work well with such abstract concepts — we understand placing one thing in one place intuitively and automatically.

Notebook-first is better than tagging-first, in my opinion, mostly because it stays out of the way. It doesn’t try to automate and encroach upon the deeply intuitive act of making connections and seeing patterns. P.A.R.A. on its own is a notebook-first system.

But if we stopped there, it would still be woefully inadequate for an economy based on creative output. As the tagging enthusiasts correctly point out, notebooks and folders actually suppress the serendipity and randomness that is at the heart of a creative lifestyle.

re. Tagging-first vs. Notebook-first

I propose a way to break the impasse (ed. between demerits of both systems): a note-first approach.

每天早上起牀後立刻着手完成最需要高度智力的複雜工作、進行判斷和決策工作

一天的一餐後的剩下的時間才適合攝入資訊

The two calendar components – the month and week indicators – represent time-varying contextual information which is shared across the individuals within a given cohort. In addition, in this example, we include also an individual time-invariant covariate (gender) and a time-varying, individual-level covariate (marketing appeals)

The Single Most Important Thing to Change Your Life: 早上起來立刻接受強光照射2-10分鐘

按键开关（Push switch

Push swi

Designing discoverable notes

A note-first approach to knowledge management means we have to think about design. You are, in a very real sense, designing a product for a demanding customer — Future You.

Future You doesn’t necessarily trust that everything Past You put into your notes is valuable. Future You is impatient and skeptical, demanding proof upfront that the time they spend reviewing notes will be worthwhile. You’ve gotta “sell them” on the idea of reviewing a given note, including all the stages any salesperson has to master: gaining attention, inspiring interest, establishing credibility, stoking desire, and making a case for action NOW.

In the case of notes, I believe the two priorities we are trying to balance are discoverability and understanding.

Making a note discoverable involves making it small, simple, and easy to digest. We accomplish this using compression: creating highly condensed summaries, without all the fluff.

But we also want to make our notes understandable. This involves including all the context: the details, the examples, and cited sources to be sure nothing falls through the cracks.

Let’s look at some examples.

If we compress a note too much, in other words, we make a summary that is too brief, we lose the context and it loses all meaning. In the note above, for example, the information it contains is highly discoverable — I can get the gist of it with just a glance.

But if I come across this note a year from now, I’ll have no idea what it means or why it’s important. It’s too compressed.

But we can go too far in the opposite direction too. If we make something totally understandable, in other words, if we include every little detail and bit of context, it loses its discoverability.

Progressive Summarization works in “layers” of summarization. Layer 0 is the original, full-length source text.

Layer 1 is the content that I initially bring into my note-taking program. I don’t have an explicit set of criteria on what to keep. I just capture anything that feels insightful, interesting, or useful.

This can include virtually any type of media, but for this article I will focus on text. There are many ways of doing this:

• Copy a paragraph of text from a PDF I’m reading, and paste it into the Evernote menu bar helper
• Type my random thoughts into a new note on the Evernote mobile app
• Dropping a Word document onto the Evernote icon in the dock on my Mac, which adds it to a note as an attachment
• Downloading all my Kindle highlights from a book using Bookcision, and then copying and pasting them into a new note
• Forward an email with useful information to my personal import address, which automatically imports the whole email to a note
• Highlight the best passages of an online article using the web highlighter Liner, which exports directly to Evernote

Layer 1 is the starting point of Progressive Summarization, like the bedrock on which everything else is built:

Layer 2 is the first round of true summarization, in which I bold only the best parts of the passages I’ve imported. Again, I have no explicit criteria. I look for keywords, key phrases, and key sentences that I feel represent the core or essence of the idea being discussed.

For Layer 3, I switch to highlighting, so I can make out the smaller number of highlighted passages among all the bolded ones. This time, I’m looking for the “best of the best,” only highlighting something if it is truly unique or valuable. And again, I’m only adding this third layer when I’m already reviewing the note anyway.

For Layer 4, I’m still summarizing, but going beyond highlighting the words of others, to recording my own. For a small number of notes that are the most insightful, I summarize layers 2 and 3 in an informal executive summary at the top of the note, restating the key points in my own words.

And finally, for a tiny minority of sources, the ones that are so powerful and exciting I want them to become part of how I think and work immediately, I remix them. After pulling them apart and dissecting them from every angle in layers 1–4, I add my own personality and creativity and turn them into something else.

Marginal distributions can now be made in R using ggside, a new ggplot2 extension. You can make linear regression with marginal distributions using histograms, densities, box plots, and more. Bonus - The side panels are super customizable for uncovering complex relationships.

Here are two examples of what you will do in this tutorial!

Not only is it okay if I don’t remember the vast majority of what I read, that actually becomes my goal. I want to offload what I’ve learned as quickly as possible, so I can forget it as quickly as possible. I want my mind to be an empty vessel, a staging ground where ideas briefly stop in their journey from the outside world, to my second brain.

Here’s the crucial thing to understand: I don’t summarize notes on any sort of schedule, in any particular order, or as a part of a workflow. I summarize them completely opportunistically, when I’m already reviewing the note for some other purpose anyway.

A given note may not be summarized until months or years after it’s been captured. Many notes may never be summarized. This is not just acceptable, it is absolutely fundamental to focusing your attention primarily on your most valuable notes. This is how your first brain works: use it or lose it. You have to be comfortable not only letting things fall through the cracks, but placing the cracks strategically so notes that don’t end up being useful automatically recede from your attention.