Merged Discussion of femr's video data analysis

Yes. But when such identification is made by the same person who makes such blatant mistakes, I have to doubt the identification, and the standard for proof is raised.
Such as ?

Then what is the problem ?

Why not have a look at the difference between static point data (camera motion) and NW corner motion....

787997325.png



Anything jump out at'ya ?

What "femr2 methods" are in play here ? I'll tell you...nowt, nada. Simply graphing the raw data from SynthEyes.
 
Last edited:
Well, please explain.
Really ?

Taking the first derivative of a series of curve fits is not the same as taking the zero'th derivative (smoothing) and then differentiating that discrete data by another method such as symmetric difference.

They are not equivalent actions.

They generate different datasets.


The very simplest explanation of which (that YOU have already had a good stab at explaining, though you did so to try and "prove femr2 wrong version the umpteenth" and perhaps missed your own point) is that if you apply symmetric difference to SG smoothed data you're actually taking a derivative of data produced by multiple curve fits, not one.

Can't find your own words on the subject. Perhaps you'd care to find them yourself.
 
Well, please explain.

More specifically, you said:
Savitzky-Golay smoothing essentially performs a local polynomial regression on a series of values to determine the smoothed value for each point. As a "bonus", derivation of each polynomial function can be easily performed to provide derivative data.
I infer from there that the derivative is performed over the polynomials, which are the ones determining the smoothed value, and therefore apply to the smoothed data, not to the unsmoothed data.

Where am I wrong there?
 
I infer from there that the derivative is performed over the polynomials
Correct, when you extract polynomial derivative order > 0.

which are the ones determining the smoothed value
Correct, when you extract polynomial derivative order = 0...simply the value at the centre-point on each curve, not a mathematical derivative at the centre of each curve function.

and therefore apply to the smoothed data, not to the unsmoothed data.
What ? The same source (unsmoothed) data is used in either case.

One outputs derivative data from the unsmoothed raw data. (order >0)

One outputs smoothed non-derivative data from the unsmoothed raw data. (order = 0)

Where am I wrong there?
It appears you are assuming a separate smoothing step prior to taking a derivative, though I can't be sure from what you're saying.

Perhaps to male it clearer to you...if you smooth the data with SG, and then use SG to generate a derivative...the derivative will be different to that obtained by generating a derivative from the original unsmoothed data, of course, as it's a different dataset... the curve fits to smoothed data will be different.
 
Last edited:
It appears you are assuming a separate smoothing step prior to taking a derivative, though I can't be sure from what you're saying.
The polynomial fit is the smoothing step.

The value of the polynomial at the point of interest is the output of the zero-order derivative.

The slope of the polynomial at the point of interest is the output of the 1st-order derivative.

Therefore, the derivative is performed over the smoothing step.

Prove me wrong. Take a velocity dataset and smooth it with the same method you used to derive your acceleration graph. Overlay the smoothed data and the acceleration graph, zooming over a zone rich in slope changes. And show that the visual slope of the smoothed velocity does not correspond with the value of the acceleration. Then I will concede I was wrong.
 
The polynomial fit is the smoothing step.
Kinda, yes.

The value of the polynomial at the point of interest is the output of the zero-order derivative.
Correct.

The slope of the polynomial at the point of interest is the output of the 1st-order derivative.
Correct.

Therefore, the derivative is performed over the smoothing step.
The source data for both is the same identical unsmoothed dataset.

Prove me wrong.
I don't need to prove you wrong, the answer is obvious.

Take a velocity dataset and smooth it with the same method you used to derive your acceleration graph.
:) Why don't you start with unsmoothed (displacement) data and do it yourself.

Overlay the smoothed data and the acceleration graph, zooming over a zone rich in slope changes. And show that the visual slope of the smoothed velocity does not correspond with the value of the acceleration. Then I will concede I was wrong.
I suggest you do these things yourself.

Also, that's not "where you're wrong".

Where you're wrong is asserting that a smoothed dataset resulting from applying SG with derivative order 0 is the source data for a dataset resulting from applying SG with derivative order >0 to the same original unsmoothed source dataset.

Perhaps to male it clearer to you...if you smooth the data with SG, and then use SG to generate a derivative from that smoothed data...the derivative data will be different to that obtained by generating a derivative from the original unsmoothed data, of course, as it's a different dataset... the curve fits to smoothed data will be different to those generated from the unsmoothed data.
 
Last edited:
The source data for both is the same identical unsmoothed dataset.
Yes. But the operation done is not merely a(n approximation of the) derivative. It includes a smoothing step.


I don't need to prove you wrong, the answer is obvious.
That is true, the answer is obvious.


:) Why don't you start with unsmoothed (displacement) data and do it yourself.
Because I don't have a SG filter that outputs derivatives.


Also, that's not "where you're wrong".

Where you're wrong is asserting that a smoothed dataset resulting from applying SG with derivative order 0 is the source data for a dataset resulting from applying SG with derivative order >0 ... to the same original source data.
I'm not claiming such thing.

I'm claiming that the first-order derivative output of SG smoothing matches visually the mathematical derivative of the zero-order derivative output of SG smoothing, and can therefore be claimed to be the derivative of the smoothed data.


Perhaps to male it clearer to you...if you smooth the data with SG, and then use SG to generate a derivative smoovative from that smoothed data...the derivative smoovative will be different to that obtained by generating a derivative smoovative from the original unsmoothed data, of course, as it's a different dataset... the curve fits to smoothed data will be different to those generated from the unsmoothed data.
Fixed that for you. I've called "smoovative" the output of SG first-order derivative, which is smoothing + derivative.

Not sure where you're going with that word salad. Let me point out, however, that the derivative obtained by central differencing applied to the SG-smoothed data, and the SG smoothing of the central-differencing derivative of the raw data, do match, as I showed in a previous post.
 
Last edited:
Yes. But the operation done is not merely a(n approximation of the) derivative. It includes a smoothing step.
All methods of generating a derivative from discrete data "include a smoothing step", including symmetric difference. The only real difference is that SD uses a linear "curve", whereas SG uses a polynomial "curve".

Because I don't have a SG filter that outputs derivatives.
Indeed :(

I'm not claiming such thing.
You said...
I am suggesting that SG shows derived velocity of the post-smoothing data, data which you have not ever shown (in order to do so, you just would have to apply SG with the same parameters and request the smoothed data instead of its derivative)
Any suggestion of pulling out smoothed (order 0) data that was not used is silly. Anyway...

I'm claiming that the first-order derivative output of SG smoothing matches visually the mathematical derivative of the zero-order derivative output of SG smoothing, and can therefore be claimed to be the derivative of the smoothed data.
Claim whatever you please. Justify your claim by doing it.

Fixed that for you.
Incorrect. I used the correct term. You've just made one up :rolleyes:

I've called "smoovative" the output of SG first-order derivative, which is smoothing + derivative.
Would you apply the same to symmetric difference ? Careful now, and don't forget to answer ;)

Making up terms is not a great idea. It blurs the edges of where you're losing yourself even more... :rolleyes:

Not sure where you're going with that word salad.
I guess you don't understand yet. I'm not sure how much simpler I can make it for you. Step one is to remove your ridiculous made-up "smoovative" term, and reinstate my original words. Then read it slowly and repeatedly until you understand.

Let me point out, however, that the derivative obtained by central differencing applied to the SG-smoothed data, and the SG smoothing of the central-differencing derivative of the raw data, do match, as I showed in a previous post.
So ? As has been said before, both here and externally...

The Savitsky-Golay method does NOT preserve maxima/minima and zero crossings when going to derivatives.


Perhaps I'll perform derivative generation for the cam#3 data, and you can inspect the results yourself. I'm most certainly not going back two YEARS to regenerate any more different views on the Dan Rather data. It's totally obvious that multiple methods result in the same general trend. I'm not at all interested in lower levels of detail, however much you'd like to quibble with femr2 until the blue sky changes colour.

If you want to explore SG derivatives, use software capable of doing it. I provided a number of SG source-code implementations a few days ago. Many others are freely available. It will be a good learning experience for you.
 
Last edited:
Let me point out, however, that the derivative obtained by central differencing applied to the SG-smoothed data, and the SG smoothing of the central-differencing derivative of the raw data, do match, as I showed in a previous post.

As I said earlier, I suggest you change the vertical axis scales such that the finer details of both your velocity and acceleration trend plots are comparable, allowing a more useful lookie-see.

I would also add, before going down this ping-pong road too much further, that whilst it is correct to say "the Savitsky-Golay method does not preserve maxima/minima and zero crossings when going to derivatives" it's also more than capable of actually doing so. It rather depends upon the source data, parameter selection, choice of base data for each derivative, ... call it a tradeoff between noise/data retention and smoothing level ;)

Still looking forward to Oystein's "god data" so that all these methods can be tested here with known underlying behaviour.

Assuming that results in superbly accurate velocity and acceleration detail I'd still be cautious of making assertions about trend behaviour for WTC7 beyond what has already been stated.

Are you aware of what I've actually asserted about WTC7 trend acceleration behaviour ?
 
Last edited:
I challenge his explanation of how he achieved his results....
Good for you. My primary interest is not on his explanations rather my focus is on the results.
...I do not say his results are false; merely unproven.
....to be precise "unproven to you". I am interested if anyone can show that femr's methods have process errors which limit the usefulness of his data and motion findings. Hence my lengthy and repeated explanations in recent posts. I'm disappointed somewhat that two of the better posters here don't want to pick up the challenge.
....The EPH fall is early motion, with early motion defined as "movement some [unspecified] time before release of the façade". It happened. Therefore there was early motion, by that definition...
Yes - but our focus in on NW corner - of a building which collapsed "shortly" after - My reasons for finding that there must have been "movement before release - of the NW Corner" explained as much as I need to in earlier posts.
... And that has been well known and thoroughly discussed for ages.
I have been clear - I am not discussing opinions rather attempting to approach the underlying topic from objective facts.
Whether more early motion can be visible in the NW corner is what I am challenging at this stage...
Change of focus - you were querying femr's explanation of his measurements because the explanation doesn't satisfy you. I'm not even referring to femr's data for what I claim about "there must have been early motion". (Of the NW corner before I get derailed back to EPH :)) There are three related aspects:
1) Was there early motion before release? - where I say "yes - must have been except for one plausible explanation which fails for other reasons - therefore must have been".
2) Can anyone detect it in terms of magnitude and time of becoming detectable? Where I think NIST did but not to the level of accuracy which prima facie applies to femr's researching.
3) Has femr detected it accurately - where "accuracy" is the topic I have been trying to get discussion of as a separate issue from whether some internet posters may not be convinced by femr's explanations. (Yes - it is a subset of "2)")
...You're failing at a very basic logic premise...
I'm still having fun - slightly frustrated at my main points being ignored...but...that's internet forumming. :o On this one no comment is needed. It's your strawman not my argument.
...We're in a very simple logic case in essence: "femr2 methods imply early motion"...
Actually, for what you were trying to say it would be "you're" - not "we're" and I think it should have been "I'm" coz I'm not confused. (I'll let you sort out which "I" is which. :D)

....When the premise is false, we don't know anything about the conclusion. It doesn't mean the conclusion is true. But it doesn't mean it's false either. It is merely unproven....
Either you or I may have needed the "Logic 101" bit. I'll pass. :)
...Nor did I say it was deliberate. But given the absence of an explanation or a quotation, sorry but your credibility has plummeted for me....
Sorry for that. I actually thought that our mutual respect was stronger than that.
Your comment as perceived was a snide comment (as was the "only read the last post" comment) and you have made no effort to clarify it...
And I have not the slightest intention of entering into a derail of personalised attack accusations to justify the objective basis for the comments. So the rest of this personalised matter gets no response. If that looks like I'm not able to rebut the stuff - tough. My SOP's are well established:
In the lack of a clarification I have to assume you have double standards, in that you can bite but not be bitten. You criticize double standards and snide comments in others and ask for responses to your criticism but won't respond to similar criticism made against you. Not to mention that I've been addressing your questions as well.
The distinction I make is between discussion of objective issues of fact and personal criticism. Where my early posts on any topic may be wrong to some extent - so I respect and respond to reasoned discussion of the areas where I may have made errors of objective fact. I rarely respond to personal commentary other than as I have done here by pointing put where objective fact ends and personal criticism takes over.

And, again, I commend your efforts in researching the issues in the topic of this thread. I'm somewhat disappointed that you don't wish to follow through with my suggestions. So, with some sadness, I'll leave you with femr. It could have been an interesting debate. :o
 
pgimeno,

No response about early motion yet...

The most direct visual "proof" of early motion is, imo...

787997325.png


To simplify...

a) Until t=60 static region motion and west edge motion are very closely matched = no motion of west edge

b) After t=60 west edge moves more than static region = motion of west edge

c) The difference between static region (camera) motion and west edge motion is the actual motion of the west edge


Significant motion of the West edge of WTC7 begins at t=60, which is ~2 minutes prior to release.


Do you have any issues with this pretty darn clear "justification" ?
 
Are you aware of what I've actually asserted about WTC7 trend acceleration behaviour ?

No response yet I see.

Well here y'are, have done the quick search for ya...

My acceleration graph shows:

a) Rapid increase in acceleration from release to somewhat over-g in approximately 1s.

At the end of this period, the NW corner had descended ~9ft

b) Slow reduction in acceleration to approximately g over approximately 1.5s.

At the end of this period, the NW corner had descended ~83ft

c) More rapid reduction in acceleration to roughly constant velocity over approximately 2s.

At the end of this period, the NW corner had descended ~270ft

Thought I'd have a peek at your acceleration trend plot...

855646088.png


...and see if the same assertions hold.

Whaddaya know, my assertions "work" for your plot too ;)
 
My acceleration graph shows:

a) Rapid increase in acceleration from release to somewhat over-g in approximately 1s.

At the end of this period, the NW corner had descended ~9ft

b) Slow reduction in acceleration to approximately g over approximately 1.5s.

At the end of this period, the NW corner had descended ~83ft

c) More rapid reduction in acceleration to roughly constant velocity over approximately 2s.

At the end of this period, the NW corner had descended ~270ft

Thought I'd have a peek at your acceleration trend plot...

[QIMG]http://femr2.ucoz.com/_ph/7/855646088.png[/QIMG]

...and see if the same assertions hold.

Whaddaya know, my assertions "work" for your plot too ;)
femr2 is over-interpreting both plots.

I can't comment on femr2's alleged Savitzky-Golay plot, because (so far as I know) he still hasn't described his methods and parameters with enough detail for others to reproduce that plot, and reproducibility is critical for checking femr2's interpretation above and the alternative interpretation I'm going to state in the next paragraph. I can, however, comment upon the true Savitzky-Golay acceleration curves.

The Savitzky-Golay acceleration curves are consistent with a) an increase in acceleration from 0 to approximately g in considerably less than 1 second, b) an essentially instantaneous change in acceleration from over-g to approximately g, and c) slightly less rapid reduction in downward acceleration from approximately g to a brief period of upward acceleration, with that reduction extending over approximately 2.5 to 3 seconds.

Please note that I have said "are consistent with". I'm giving an example of an alternative interpretation of the data that is at least as likely as femr2's interpretation, quoted above. (I will provide evidence for its likelihood in part 3 of my continuing series on Savitzky-Golay smoothing.) It's quite probable that the true acceleration was different from both femr2's interpretation and the interpretation I have just stated.

Please note also that I am not suggesting this alternative to femr2's interpretation is significant in any way. Before I could argue for the significance of an interpretation, I'd have to have some purpose in mind for which the interpretation might signify something. So far as I know, femr2's video analyses have had no purpose, and my own analysis is primarily pedagogical.
 
No response yet I see.
Busy here. Patience.

Whaddaya know, my assertions "work" for your plot too ;)
Duh. I was trying to reproduce your results in order to make a point about your method, not to do my own analysis of the data. Hardly surprising that your assertions "work" on a graph which was deliberately created to be as close to yours as I could.

But I'm out of that game. If you don't want to publish your methods, your results are unverifiable and thus useless.
 
Duh. I was trying to reproduce your results in order to make a point about your method, not to do my own analysis of the data.
Just to make that point more obvious, a quote:

So, armed with that tool I went ahead and started experimenting. I was able to reproduce a graph with a much more similar trend to that of femr2's, by using SG with a window of 51 and polynomial degrees 1 and 2, instead of the same for each pass. Here's a comparison for reference:

[qimg]http://www.formauri.es/personal/pgimeno/xfiles/11-s/femr2-qtiplot-comparison.png[/qimg]
 
Duh. I was trying to reproduce your results in order to make a point about your method
Yet another case of "prove femr2 wrong version the umpteenth" :rolleyes:

Am going to have to make some "femr2" smileys :whistling

not to do my own analysis of the data
Of course not. What would be the point in that ?! :rolleyes:

Hardly surprising that your assertions "work" on a graph which was deliberately created to be as close to yours as I could.
Try W.D.Clinger's...
32653665.png


...or even tfk's "effort", though I don't think that one is particularly useful...

51514794.png



But I'm out of that game.
What a surprise.

If you don't want to publish your methods
ROFL. Exact procedure for one graph produced nearly two YEARS ago has been witheld, and the "response" has been highly entertaining.

All tracing method details have been published many times.

All derivation method details, from the old polynomial fits, symmetric difference plots, use of Savutzky-Golay smoothing, ...have all been published many times.

There's even a 1-pass SG plot released only a week or so ago...
621711303.png

...and...
120219215.png

...for which full details on reproduction have been provided.

You've simply run out of valid "criticism". Another failed instance of "prove femr2 wrong version the umpteenth". :rolleyes:

your results are unverifiable
ROFL. Try the graphs above. Oh, wait, you can't, as your implementation of Savitzky-Golay that you found in your filing system doesn't output SG derivatives :rolleyes:

and thus useless.
Of course, no other conclusion would possibly be expected :rolleyes:
 
The Savitzky-Golay acceleration curves are consistent with a) an increase in acceleration from 0 to approximately g in considerably less than 1 second,
Your 1-pass SG profile has been affected by very noisy near-release data (between 11s and 12s). Suggest you give it a bit of smoothing somewhere along your data processing chain ;)

b) an essentially instantaneous change in acceleration from over-g to approximately g,
Instantaneous ? :jaw-dropp

You would'nt be assuming SG has spread the width of the "over-g period" would you ?

and c) slightly less rapid reduction in downward acceleration from approximately g to a brief period of upward acceleration, with that reduction extending over approximately 2.5 to 3 seconds.
Don't you mean deceleration ? Increase in rate of velocity reduction...

I will provide evidence for its likelihood in part 3 of my continuing series on Savitzky-Golay smoothing.
I can hardly wait.

It's quite probable that the true acceleration was different from both femr2's interpretation and the interpretation I have just stated.
Then get some "better" trace data and produce some "better" results :)

Please note also that I am not suggesting this alternative to femr2's interpretation is significant in any way.
But it "disagrees with femr2", which is awesome. Yay.

Before I could argue for the significance of an interpretation, I'd have to have some purpose in mind for which the interpretation might signify something. So far as I know, femr2's video analyses have had no purpose, and my own analysis is primarily pedagogical.
As long as you're learning something, that's great. Can't say your public reputation has been helped, but that's not my doing.
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom