datageist

@Fat32 I think this answer explains it pretty clearly. dsp.stackexchange.com/a/9512/35. It should be clear how to experiment with the designs in Matlab after going through it.

@Fat32. Yes, it's extremely surprising, and that's part of the reason people got so excited about them :). Check the two answers in the comment I posted to MattL (one of them has source code to do the design using numpy).

@Fat32 Moving to chat doesn't mean they'll be removed--just transplanted to a setting that's more suitable for discussion. The short answer for SG filters, is that the local polynomial fit is dependent on the sample spacing, which happens to be FIR for equally-spaced samples. The filter coefficients serve to evaluate the least-squares polynomial at 0 using a dot product with the signal values.

@MattL. I really should move all these comments to chat, but these answers may save you the trouble of explaining (dsp.stackexchange.com/a/9494/35, dsp.stackexchange.com/a/9512/35).

The Stanford Fourier Analysis lectures are amazing as well.

I learned more heavy linear algebra from watching a few of those lectures than any of the MIT OCW stuff.

It's like linear algebra combined with differential equations.

Interesting. The one I really want them to put in coursera format is Linear Dynamical Systems.

Yeah, I was really looking forward to that one. Maybe next time around. Like what are the pre-reqs for it?

How is the PGM class so far?

Another trick I like is finding graduate classes with a module that reviews some prerequisite. Those "reviews" often teach the subject better than an entire class.

That way one author will hopefully fill in the gaps left by the others.

In the absence of good lectures, that is.

Well, learning just about any difficult subject tends to require finding at least 3 really good books from different authors.

Yeah, exactly. Or say controversial things like "nobody ever uses this".

And that's the real value of the all OCW stuff (for me at least).

In other words, there's lot of ambient wisdom that gets conveyed during lectures that most people would never put into a book.

The main problem with books (in relation to self-study) is that they are too formal.

Yeah, I like to verify most of my knowledge with programming (whenever possible), which I find makes it stick more than exercises. Unless the exercises happen to be really well designed...

Yeah, I've got mixed feelings about these online courses compared to how I learn on my own.

Ha :)

Deadlines provide a minor motivation, but good content is better.

Well, deadlines are OK if they're relative rather than absolute.

There's too much of an effort to re-create a "classroom" experience, and it doesn't transfer.

Which just seems capricious and arbitrary. Online courses should be asynchronous.

Yeah, even when I took the Machine Learning one, I started it almost 2 months late, so even though I technically got a perfect score, they wouldn't grade it that way.

The nice thing about Coursera is that all the content was contained in the lectures.

Yeah, this MIT one is wearing me out. I was signed up for 3 coursera courses when this one started, and had to drop out of all of them :)

Yep.

I think future verisons of the course will be better--this is kind of the prototype.

The infrastructure is more coherent/mature than the content.

Yeah, they've really got some manpower behind it, and some really good programmers.

Interesting. I haven't tried anything from Udacity yet.

They have a full circuit simulator in the browser, for example.

The interesting thing about MITx vs. Coursera is that the infrastructure is more comprehensive.

Right, that's the main difference--prerequisites are assumed. Diff eq's, some electronics background, etc.

The lectures felt coherent.

It's supposed to be like a "real" MIT class, so it's more immersive. I actually enjoyed watching the OCW circuits lectures more than I'm enjoying this class, if that makes sense.

They've got one professor doing the lectures, and another doing the assignments, and the content doesn't seem to connect 100%.

Much more time consuming than the coursera stuff was, mainly because they don't pull any punches, technically speaking. However, the coursera stuff felt MUCH more coherent.

I went through the first version of Machine Learning though. Pretty good.

I was signed up for the PGM class, but the MITx circuits one is taking up all my ambient time at the moment.