31 August 2020

Parallax View - part the twenty first

Another week's up, so here are today's numbers:
  >= 1,000 - 269
100 to 999 - 1,276 

(New) grand total of counties reads at 3,090. While I've not gotten into an argument with the Topo folks over this re-definition of 'affected' counties, it certainly looks like a move to bolster ex-President AuH2O 2020's perpetual lying.

In an early April missive, I mused that the 'early' data and historical data with other coronaviruses implied that herd immunity was going to be an iffy proposition. Now, we have new reporting that Covid-19 immunity (in the individual) may be transient, at best. Batshit J. Moron spent his week acting like Covid-19 is in the rearview mirror. Of course he did; to do otherwise explicitly admits that he and his elves have screwed the pooch. I doubt that Mother Nature did anything other than flip him a Cosmic Bird. If there is a God, those almost 2,000+ nutballs on the South Lawn will all die of Covid. Before they can vote.
What caught experts' attention about the case of the 25-year-old Reno man was not that he appears to have contracted SARS-CoV-2 (the name of the virus that causes Covid-19) a second time. Rather, it's that his second bout was more serious than his first.

More to the point is whether letting Covid-19 run rampant (aka, Sweden) is a viable approach, aka enough folks get sick means that herd immunity is achieved before half the population is really, really sick or dead. If individual immunity is too short to remain effective among the earliest infected simply means that community immunity must plateau at the number in the community that can be infected and recover (the dead don't help gaining herd immunity; suffering with no benefit) within that timeframe, say three or four months. Which implies that there will never be enough time to reach herd level of immunity. There exists no evidence, that I've seen, that herd immunity level can be attained in such a short period. Unless you round up the lower classes, lock them in cattle cars, spray Covid, and hope that will be enough to reach herd. All without bothering the privileged.
But then, some 48 days later, the man started experiencing headaches, cough, and other symptoms again. Eventually, he became so sick that he had to be hospitalized and was found to have pneumonia.

So, a confirmed case with immunity kaput at two and half months. If, and that's a significant if, this patient is the norm, then the 'let them all get sick and we'll all be safe soon enough' is bullshit. With a batshit leader, I guess that's to be expected.

What none of the reporting I've seen has addressed: is re-infection due to lost immunity to Covid-19, or is it due to sufficient mutation of initial Covid-19 to make it less, or even in-, visible to the immune system? Confounding that is the contrary reporting that those who've been minimally affected by other coronaviruses, aka the common cold, may have some immunity to Covid-19.
In labs all over the world lately, scientists working on COVID-19 have stumbled on an intriguing sort of finding again and again. They've found that blood samples from healthy people who were never exposed to the SARS-CoV-2 coronavirus contain reactive immune cells and targeted antibodies that could, perhaps, help stave off COVID-19.

Since my cold experience runs thus: onset is late afternoon, go to bed, stay in bed the next day, get out of bed on the third day; and I'm done with it. For me, Covid-19 appears to be a non-event. Or so one hopes.

Conflicting evidence. That's what makes science, unlike nutballs who assume they already, and forever, know everything. Bleach my hair. Bleach my skin (white is always better). Bleach my lungs.

White Privilege?

So, you say White Privilege is just lower-class complaining? It's been a few weeks for the story to shake out, but the facts that have emerged is that the UK's draconian system of college entrance can be systematically biased. To be fair, sort of, this outcome came from a data science approach in lieu of regular exams. AI, to paraphrase the fellow running the Shoah Project's interactive program (re-run last night on "60 Minutes"), is mostly artificial and not very intelligent.

"The Guardian" has published a number of reports, so this is just one among many.
That tells us already that the history of the school is very important to Ofqual. The grades other pupils got in previous years is a huge determinant to the grades this year's pupils were given in 2020. The regulator argues this is a plausible assumption but for many students it is also an intrinsically unfair one: the grades they are given are decided by the ability of pupils they may have never met.
[my emphasis]

A cute way to institute class (read: White) privilege:
The decision to give small classes the ability to receive their teachers' recommended grades is not in the algorithm but led to a boost for elite private schools.

The choice to take the results of the algorithm and further tweak the grade boundaries to prevent overall grade inflation is not in the algorithm but further depressed the larger classes in favour of the smaller.
[my emphasis]

If ever one needed an indictment of Bayes, this be it. Mammy Yokum has spoken.

26 August 2020

Parallax View - part the twentieth dot five

Another week's up, so here are today's numbers:
  >= 1,000 - 286
100 to 999 - 1,264

Insanity reigns supreme, thus an intra-week update.

(New) grand total of counties reads at 3,084. While I've not gotten into an argument with the Topo folks over this re-definition of 'affected' counties, it certainly looks like a move to bolster ex-President AuH2O 2020's perpetual lying.

[Tuesday Morning, in the event that deadline addicted Real Media find it before next Monday (they sure ought to)]
If FDA can fudge data so blatantly, why are we to believe that the 'decrease' in Covid is anything but manipulation by Batshit J. Moron's ringer Thiel doing the data collection and reporting?
The publicly released data from the Mayo Clinic shows that, among the larger group of more than 35,000 patients, when plasma was given within three day of diagnosis, the death rate was about 22 percent, compared with 27 percent when it was given four or more days after diagnosis.

Of course, we've no idea what the death rate was for a comparable (that word again) sample who got either SoC or placebo. This recent report puts a similar (best I'll assert) rate at 20.3%.
All-cause death rates were 20.3% overall, 34.7% among ICU patients, and 38.1% among those receiving mechanical ventilation. The death rate among white patients was 23.1%, while it was 19.2% among black patients.

So, plasma is what ex-President AuH2O 2020 said: "This is a powerful therapy... a very historic breakthrough." Yeah, right. Plasma has been used with many diseases for more than a century.
The first valid trial was done in 1892 with diphtheria using blood serum initially.

Or, to be blunt:
"For the first time ever, I feel like official people in communications and people at the F.D.A. grossly misrepresented data about a therapy," said Dr. Walid Gellad, who leads the Center for Pharmaceutical Policy and Prescribing at the University of Pittsburgh.

So, tell me: why should anyone believe the Miracles (made from ginned up, and even corrupted, data) coming from Dear Leader Yo! Semite of Thigh Land and his minions? I'll take malice over incompetence with this bunch every time.

[Wednesday morning, aka I Told You So]
So not only does Hahn suck Batshit J. Moron's dick, but so does Azar. Which latter is less of a surprise than Hahn. Azar was known to be an acolyte from the get go.
The CDC changed the site on Monday. Here's what it says now: "If you have been in close contact (within 6 feet) of a person with a COVID-19 infection for at least 15 minutes but do not have symptoms, you do not necessarily need a test unless you are a vulnerable individual or your health care provider or State or local public health officials recommend you take one."

Or, as Dear Leader Yo! Semite of Thigh Land put it:
If we stop testing right now, we'd have very few cases, if any.
-- ex-President AuH2O 2020/2020

Ya got shit data, ya make shit decisions. But that's the point, is it not? A Feature, not a Bug. Not a subtle sort of guy, now is he?

[Wednesday afternoon, aka That Didn't Take Long]
A sudden change in federal guidelines on coronavirus testing came this week as a result of pressure from the upper ranks of the Trump administration, a federal health official close to the process tells CNN.

"It's coming from the top down," the official said of the new directive from the Centers for Disease Control and Prevention.

I suppose Batshit J. Moron will set the Stasi out to find the 'leaker'.
After we got off the phone, Whipple texted a final thought: "A good chief of staff knows that the best way to prevent damaging leaks is to stop doing illegal, stupid stuff. You don't have to be James Baker to figure that out."

24 August 2020

Parallax View - part the twentieth

Another week's up, so here are today's numbers:
  >= 1,000 - 286
100 to 999 - 1,264

(New) grand total of counties reads at 3,084. While I've not gotten into an argument with the Topo folks over this re-definition of 'affected' counties, it certainly looks like a move to bolster ex-President AuH2O 2020's perpetual lying.

The Kiddie Korps has been at the back of the bus, smoking weed, during this catastrophe, on the whole. Then this report comes walking by.
For your sake, the sake of our community, and for continuing our semester on campus, please observe health protocols," [Norte Dame university president Rev. John I. Jenkins] said. "Avoid behavior that puts yourself or others at risk.

Today, we find that more and more schools are being infected.
Universities in at least 19 states have reported outbreaks, despite health protocols on campus. Many outbreaks are tied to large group gatherings like parties, leading some schools to suspend students and organizations for breaking social distancing rules on and off campus.

Will we ever find out the number of cases?
Many of these gatekeepers have pointed to medical and educational privacy laws as reasons to withhold even basic counts of coronavirus cases. That's despite federal guidance saying those laws aren't barriers to disclosure and legal experts who note that schools can share information as long as they don't identify individuals.

If you don't test and you don't report, 'it disappears like a miracle'.

Was on the verge of posting, when I made a belated run through today's NYT. And found this report on sports and covid. Yikes!!! Is all I can say.
Then, by early August, shortly after the German study was published, examples began to surface of otherwise healthy athletes showing signs of myocarditis after they recovered from Covid-19. Boston Red Sox pitcher Eduardo Rodriguez was sidelined for the season. The Indiana freshman lineman Brady Feeney's doctor was concerned about Feeney's heart after a difficult bout with the virus. A 27-year-old former Florida State basketball player, Michael Ojo, who had recovered from the virus, died of an apparent heart attack at a practice in Serbia.
[my emphasis]
Much of the reporting is about the cover-up of college athletes' experience with Covid. The reporting states that some studies show up to 15% of recovered Covid patients with myocarditis. How many of them have to die before the cover-ups end?

More and more this looks like ex-President AuH2O 2020 is pushing a massive cover-up. The buckling of FDA over plasma is just another brick in the wall.
Dr. Paul Offit, director of the Vaccine Education Center at Children's Hospital of Philadelphia, said he thought it likely the White House pressured the FDA into pushing through the EUA.

"I think what's happening here is you're seeing bullying, at least at the highest level of the FDA, and I'm sure that there are people at the FDA right now who are the workers there that are as upset about this as I am," Offit told CNN's Wolf Blitzer.

HCQ all over again. Or, in the words of Maryanne Trump, "It's the phoniness of it all. It's the phoniness and this cruelty. Donald is cruel."

21 August 2020

One Classy Dame - part the sixth

There are two issues/problems, and one contra, with using PMEM/Optane to the fullest extent:
1 - is it net efficient to update-in-place from cpu cache lines to Optane, rather than through all the caches, buffers, and the usual memory hierarchy used with conventional HDD/SSD machines?
2 - is it possible to update-in-place with Optane and implement ACID conditions using only memory latches; aka no traditional locks?
3 - after all, RDBMS engines update-in-place their buffers, so why not PMEM?

The notion, mentioned here in terms of AS/400 (actually first released years earlier on S/38) single level storage (with database object, rather than file protocol) and TI-990 memory-to-memory execution, that RDBMS (or any transactional application for that matter) would be a major beneficiary of persistent memory if it could be written to while skipping past all those caches and buffers and non-key data that exist in the standard compute architecture. One might envision a new cpu architecture designed specifically to persistent memory, rather than volatile memory + durable storage. But that's well into the future, if ever. What can be done today with Optane and the X86 ISA? Turns out, some pretty neat stupid pet tricks.

To wit, found a paper from Intel they released last fall which takes a detailed look at what can be done. In a previous essay, One Classy Dame - part the second, the Oracle implmentation (of unknown numbers in the wild) was discussed.

The Performance section of this Intel paper has this caveat:
Due to persistent memory's low latencies—as compared against traditional HDDs and SSDs— and the lack of a queue of requests in the device (see Figure 1), only synchronous I/O is recommended. Doing I/O in an asynchronous fashion adds extra software overhead and, as we have already discussed above and will see in the performance numbers shown below, persistent memory shifts the I/O bottleneck from hardware to software. If your application can't wait for an I/O operation to complete even at low latencies, then asynchronous I/O makes sense. In that case, however, you may need to reevaluate if persistent memory is the right technology for your I/O needs as you will lose the media speed advantage.
[my emphasis]

In other words (mine): if you write small, fast transactions (in the pre-persistent memory world) then you'll likely get into warp speed with Optane and App Direct. On the other hand, if you write 'long lived' transactions, not so much. Moreover, in most multi-user/application databases, simultaneous users don't generally update the same rows and columns, modulo 'batch' transactions and even these are generally run in some kind of pseudo-single user mode. With more granular 'locking', the opportunities for transaction collision diminish.

That said, synchronous writes are not incompatible with ACID, of course. Reading up Weikum and Vossen, you'll see that transaction integrity is based on whether an update can be transformed into a serializable history. The notion, after all, is that these are set operations, and each appears to the user to be 'all at once' and time ordered amongst themselves. With persistent memory, hardware gets most of the way there. RDBMS have to figure a way around asynchronous writing with locks, latches, and buffer managers especially when running higher than row-level locking (or massive copying with MVCC). The specifics of how Oracle and SAP have used Optane are only sketchy.
Intel has devised adaptions for specific software such as SAP but a general application software adaptation method would be helpful. And this is where Formulus Black, a venture-backed in-memory computing vendor, comes in.

This paper from SAP provides a bit of detail.
For example, Intel and SAP engineers demonstrated how a server, equipped with DRAM only and the SAP HANA 2.0 SPS 03 platform with 6 terabytes of data, required 50 minutes to start, including data loading. By contrast, a server equipped with a combination of Intel Optane DC persistent memory and DRAM reduced the start time of the SAP HANA 2.0 SPS 03 platform by 12.5x to only 4 minutes.

Not the most useful metric, but rather concrete. Of more note
App Direct Mode is the only mode that is currently supported by SAP HANA 2.0 SPS 03. In App Direct Mode, Intel Optane DC persistent memory modules appear as byte-addressable memory resources that are controlled by SAP HANA 2.0 SPS 03. In this mode, the persistent memory space is controlled directly by SAP HANA.
Note that SAP HANA uses persistent memory for all data that resides in the column data store.
[my emphasis]

In other words, code is in DRAM while data is in Optane, and may be update-in-place.

Nor have I found any report laying out why direct writing from cpu cache lines to persistent memory can't be done. Just from the Oracle documents and the SAP report just above, that's what appears to be going on. It would seem that, from a transaction point of view, it is only necessary that there be a start-log-entry record, then the data, then an end-log-entry record which happens in the log (which is not some sequential write store, either) in persistent memory, too. That log record would have the engine's unique row identifier (never visible to the user, btw), a timestamp/transID, column identifiers, and new values for said columns. Then, for updates, latch the 'old' data which might just be the modified columns, change the column values in place to 'new' data, release the latch on the log record, and release the latch on the 'old', now 'new' data. For an Insert, once committed, simply adding a pointer to what was originally the 'log' record of the Insert should suffice. Log records for committed data are not needed. Now, what kind of logging is this (noting that logging, in the minds of many, still thinks in sequential terms, since that was the fastest write to HDD)? In this scenario, a log record can be anywhere in the persistent memory patch, carrying only the necessary flags to mark it as log record. Recall that RDBMS logging is solely for integrity, not application history, so once the 'new' data is committed/durable, the log image is of no use. Now, I'll suppose that there are myriad applications written on RDBMS that ignore the mantra of 'small transactions done really fast' and thus depend on large logs generated from 'long lived' transactions for crash recovery. Ain't no help for those dumb as a sack of hair.

Unlike common row locking (and MVCC), the whole row wouldn't be replicated. Moreover, non-key columns could be updated 'at will' by any writer. Even in these olden days, last writer wins. That would still be true, just faster; there being no locks to contend with, it's possible (I think, don't ask me for details) with persistent memory to latch only key/unique columns on a row. Now that would speed things up considerably.

Here is a very old report (aka, not bleeding edge ideas these days) on memory resident RDBMS, from IBM, of all places.
A latch operation typically involves far fewer instructions than a lock operation, as a latch's data structures are statically allocated and directly addressable. In fact, in R* [Williams 821, a distributed relational database system prototype developed at IBM research, a latch and unlatch operation used about 20 CISC (IBM 370) instructions, which was roughly an order of magnitude less than a lock and unlock operation [Yost 92, Gray 89, Lehman 891. Thus, given a latch's relatively short pathlength, one might dismiss the concern of latches imposing a significant overhead in the overall storage component pathlength.

So, one might expect that such dubious applications wouldn't implement App Direct use of Optane, simply use Optane SSD as a faster, well, SSD. That won't buy much, compared to re-engineering using Optane as persistent memory.

17 August 2020

Parallax View - part the nineteenth [update the second]

Another week's up, so here are today's numbers:
>= 1,000 - 317
100 to 999 - 1,236

(New) grand total of counties reads at 3,078. While I've not gotten into an argument with the Topo folks over this re-definition of 'affected' counties, it certainly looks like a move to bolster ex-President AuH2O 2020's perpetual lying.

So, going to school is so simple. We all know that school kids are 'virtually immune' to this wee little critter. Yeah, right.
The number of quarantines since schools in Woodstock and the rest of Cherokee County opened last week is just that -- impossible to ignore. At least 478 students and teachers were asked to quarantine after possible exposure to Covid-19 in the first five days of classes. By early Tuesday, that number had swelled to 925 students and staff. Nearly one-third of the district's schools -- which teach 30,000 students from elementary to high -- have been impacted.

Georgia, out of its mind.
Back on the playground in Woodstock, Jamie Chambers said he feels sometimes like he's fighting a losing battle. In this deeply conservative part of Georgia, he says politics continue to be put before science, with even some school officials telling staff that they believe the coronavirus is a hoax, he said.

But, as usual in Hiaasen manner, Florida is truly out to lunch.
A mandate from a local sheriff in Florida is going against the recommendations from scientists and government officials: masks will not be worn.

That was the requirement spelled out in an August 11 email to the department from Sheriff Billy Woods of the Marion County Sheriff's Office.
Marion County set a single-day record on Tuesday for the most coronavirus deaths, with 13 reported, according to the Ocala Star Banner.

One thing quant doesn't do well is infer future behavior and events from newly revealed current events, the data just doesn't exist yet. In the context of this series, consider the reduction in testing and the increase in positivity in the states still on fire. ex-President AuH2O 2020 can assert that cases are lowering, when they almost certainly are not. Hospitalizations continue to rise where it matters most. The result of such a trend, should it continue, is to sweep Covid under the national rug. Folks, especially geezers and Darkies, just never get the diagnosis and die anonymously in residence. A perfect outcome.

Well, turns out UNC has thrown in the towel. The question not been voiced: what will ex-President AuH2O 2020do when the first student, be it college or younger, dies? After all, he's the one that's been insisting that kids are immune and OPEN THE SCHOOLS NOW!!! Blame that on Biden?

[update the second]
Well, now we get a 'national' correspondent asking, pretty much the same question
Any decision-makers must ask themselves: How many students, faculty, and staff are you willing to let get sick because you reopened dorms and brought students back into classrooms? How many deaths will it take until we once again close our doors? I hope we don't have to find out.

Yeah folks, how many body bags is too much?

12 August 2020

Pyrrhic Loser - part the second

Yet another series on yet another topic. ex-President AuH2O 2020 and Moscow Mitch are determined to crush wage earners under their boot heels even more than they already have. A populist, he says, with $300,000 initiation fee (boosted right after he 'won' the election) golf clubs. Right. Recall that ~$2 trillion tax cut that went almost entirely to Phat Cats? Well, they're back.

Moscow Mitch is leading the crusade to, at least, drastically reduce the supplemental $600/week Covid aid for the unemployed. The justification is that these are all layabouts who don't want to work. Ignoring the simple fact that many of these lower-middle class, blue collar businesses are direct service locations which are just too toxic to operate. We've seen what happened in the Dumb Red States that ran, lemming like (it's just a legend, but what the hell?), off the cliff by 're-opening' while infections were still stupid high. These are the states that are currently burning up with Covid, hitting and exceeding hospital beds in many small cities and counties, and really blowing out ICU support. Some of these hospitals have only single digit ICU support.

What the lovers of ex-President AuH2O 2020 and Moscow Mitch are too stupid to realize, so far, is that the supplemental is nation-wide and not modified by regional/state/MSA wage rates. In other obvious words: it's ex-President AuH2O 2020's very own zealots that are the ones making out by not working by a whole lot more than the unworthy Blue State bartenders.

Here's the data. Just click on the toggle for 'Median wage' low to high to see who gets the biggest boost. Yeah, Moscow Mitch, stick it to your base.

10 August 2020

Parallax View - part the eighteenth

Another week's up, so here are today's numbers:
  >= 1,000 - 329
100 to 999 - 1202

(New) grand total of counties reads at 3,073. While I've not gotten into an argument with the Topo folks over this re-definition of 'affected' counties, it certainly looks like a move to bolster ex-President AuH2O 2020's perpetual lying.

Among other things, there's growing evidence that 30 days is nowhere near accurate. Here's one report.
Many long-haulers say their doctors doubted their symptoms were as severe as they were saying. Roberts says her original primary care physician insisted it was just stress and suggested she watch Lifetime movies and do puzzles to calm down. "I know stress," Roberts says. "This was not stress."

This report gives specifics.
The list of lingering maladies from COVID-19 is longer and more varied than most doctors could have imagined. Ongoing problems include fatigue, a racing heartbeat, shortness of breath, achy joints, foggy thinking, a persistent loss of sense of smell, and damage to the heart, lungs, kidneys, and brain.

This is turning into the same conflict between the afflicted and physicians that exists in the Lyme disease space.
When news of COVID started spreading, I immediately took it seriously. That's not me bragging; those are just the instincts I've developed after more than a decade of living with Lyme disease. I've been told time and time again over the years that my condition makes me immunocompromised — that part of what Lyme does is it debilitates my immune system.

Many, if not most, of the medical community deny that Lyme persists beyond the treatment regimen.
However, many other figures in the medical establishment maintain that there is no such thing as chronic Lyme. Rather, the label is given to patients with perplexing long-term illnesses by willing clinicians when nothing else seems to fit.

There are a number of ways to calculate the number of affected counties for some disease. The most accurate is to track, via case files, those having the disease on a given day. With Covid, I doubt that anyone or agency is doing that. The easiest way is to take the daily new case count and add it to yesterday's. That gives you cumulative cases for the county, that's where the 5,000,000+ national number comes from. Obviously, most of the folks who were infected in March are now either recovered or dead; except of course for the long-haulers. To calculate by the new definition, you simply subtract the total found at 30, or whatever, days previous. That, naturally, assumes that no one remains affected longer than whatever duration you've chosen.

It's going to disappear. One day, it's like a miracle, it will disappear.
--ex-President AuH2O 2020/28 Feb. 2020

For whatever reason, not one reporter (that I've seen any record of) has ever challenged the fucking moron to name one disease that 'disappeared' spontaneously, without human intervention. A few days ago, a reporter did challenge him about his repeated assertion that the Veteran's Choice law was enacted 'by' him; that legislation was passed in 2014. ex-President AuH2O 2020 then walked out of the press conference. Hurt his feelings.

06 August 2020

If Not Now, When?

Intrepid reader knows that Dr. Codd's view is that transactional data must needs be controlled centrally by a transaction engine; any data fiddling on the client is just by-the-way, and not to be trusted. Yet today's java jockey continues in the steps of the COBOL cowboys of your grandpappy's day. At least in those 370 days, the 'client' was just a patch of memory on the same machine as the 'data files', so the client code could deal with CICS, or your custom transaction code. Yes, some IBM terminals of the day could be structured to execute some rudimentary editing.

Well, at least someone finally gets it, and even in a quasi-RDBMS venue:
The colossal quantity of data required to simulate this realistic world in real time is managed using Microsoft's Azure cloud-based platform. Azure does the heavy lifting, so your home computer doesn't need to be of the super-duper variety to deliver a convincing experience.
[my bold]

So, why not relegate the PC to VT-220 status (well, with pretty pixels too)? You get ironclad (to the extent that app developers care) data integrity with no additional coding, just inputting the specs into the constraints. And perhaps a trigger or two. And may be a stored proc, if that's your preferred poison.

03 August 2020

Parallax View - part the seventeenth [update]

Another week's up, so here are today's numbers:
>= 1,000 - 322
100 to 999 - 1124

Grand total of counties reads at 3,065. Something happened the middle of last week, driving down the total counties and the distribution. For myself, I see the greedy little fingers of Thiel and Palantir.

The Topo folks offer this to explain the major shift in data.
Last week, we updated our maps to show county data for "active" cases rather than "confirmed" cases. This resulted in the number of cases per county going down because:
active = any case reported in the last 30 days; cases older than 30 days are considered recovered, or may have resulted in death, and thus are not counted as "active"; this is a representation of current COVID data in a county
confirmed = any case reported at any point in time, no matter how long ago; this is a good representation of total population exposure but not of the current risk level
Utah data, on topo's maps, is built from data that is published by the New York Times. The reason we do this is because Johns Hopkins does not report Utah data for all Utah counties.

Don't Look Behind the Curtain

Long time dear reader might recall my jaundiced view of quant market trading. "Data is driven by events, not the other way round." Yet we see many stat folks going whole hog into Wall Street, dragging all manner of ever more abstruse data procs. Turns out, I'm not fully alone. (I had to switch to Chrome from Firefox to open the link; you may also.)
"It's easy to train a system to recognize cats in YouTube videos because there are millions of them," Chan says. In contrast, only a few such large swings in the market have occurred before. "You can count [these huge drops] on one hand. So it's not possible to use machine learning to learn from those signals."

In other words, events drive the data, and if there aren't enough events to drive enough data, you'll never see enough of them to drive a model. You're much better off reading 'The New York Times' and 'The Guardian' and other large dailies. That's where the Covid-19 event was reported early, and if you had a mind to, you could have put a few Bongo Bucks into a handful of micro-cap 'vaccine' developers around March 1 and cashed out in the last week or so with life altering amounts of Bongo Bucks. No AI needed.