I am trying to generate the delta-MFCCs. Apparently there are several implementations. I found the "regression" formula link here. But I don't understand why Librosa uses Savitsky-Golay filter, which is a smoothing filter. I have not found any blog/article which shows the use of Sav-Gol filter for delta mfcc. And, obviously, the deltas returned by the two don't match.
Can someone elaborate on it?
Related
When trying to upload the fasttext model (cc.nl.300.bin) in gensim I get the following error:
!wget https://dl.fbaipublicfiles.com/fasttext/vectors-crawl/cc.nl.300.bin.gz
!gunzip cc.nl.300.bin.gz
model = FastText_gensim.load_fasttext_format('cc.nl.300.bin')
model.build_vocab(cleaned_text, update=True)
AttributeError: 'FastTextTrainables' object has no attribute 'syn1neg'
The code goes wrong when building the vocab with my own dataset. The format of that dataset is all right, as I already used it to build and train other (not pre-trained) Word2Vec and FastText models.
I saw other had the same error on this blog, however their solution did not work for me: https://github.com/RaRe-Technologies/gensim/issues/2588
Also, I read somewhere that I should use 'load_facebook_model'? However I was not able to import load_facebook_model at all? Is this even a good way to solve this problem?
Any other suggestions?
Are you sure you're using the latest version of Gensim, 4.0.1, with many improvements to the FastText implementation?
And, there you will definitely want to use .load_facebook_model() to load a full .bin Facebook-format model:
https://radimrehurek.com/gensim/models/fasttext.html#gensim.models.fasttext.load_facebook_model
But also note: the post-training expansion of the vocabulary is best considered an advanced & experimental function. It may not offer any improvement on typical tasks - indeed, without careful consideration of tradeoffs & balancing influence of later traiing against earlier, it can make things worse.
A FastText model trained on a large, diverse corpus may already be able to synthesize better-than-nothing guess vectors for out-of-vocabulary words, via its subword vectors.
If there's some data with very-different words & word-senses you need to integrate, it will often be better to re-train from scratch, using an equal combination of all desired text influences. Then you'll be doing things in a standard and balanced way, without harder-to-tune and harder-to-evaluate improvised changes to usual practice.
Crossfilter supports dimensions with arrays since version 1.4.0-alpha.06 https://github.com/crossfilter/crossfilter/wiki/API-Reference#dimension_with_arrays
Is it possible to exploit this functionality in dc.js?
I haven' found any examples yet... I am aware of the method described in Is there a way to tell crossfilter to treat elements of array as separate records instead of treating whole array as single key?, but that works with earlier versions of crossfilter too (it does not make use of the new functionality).
With help from comments above, I managed to get this working, very simple indeed if you use the right version of crossfilter (1.4.0). Works fine with rowChart() and pieChart().
I made a an example bl.ock based on previous approaches:
http://bl.ocks.org/emiguevara/4bd152a8828f6b31270702d97dc0133d
I would like to visualize with aChartEngine a series of measurements. For this I have double values in which to compare the results and are currently still a string with time. Currently I use a line graph with the results and the number. I would now replace by the time the number. Unfortunately, I do not know how and find no suitable examples.
Edit
Okay, have found a good example and its work. But how i can make it flexible like the normal label?
There are plenty examples of using the AChartEngine APIs in the official demo application. See these instructions in order to figure out how to download the demo source code.
So I have no problem simplifying using topojson from the command line using the -s flag, however, I can't figure out how to do it from the node module.
I see a topojson.simplify() method, but I can't figure out how it works as there is no documentation.
Does anyone have any insight?
By looking at the simplification tests for topojson, I was able to figure out how to use toposjson.simplify(), but I can't fully claim to know whats going on. You can see the tests on the topojson github.
Basically topojson.simplify takes a topology input and has 2 possible options for simplification, "retain-proportion" and "minimum-area", you can also pass the coordinate system, aka "cartesian" or "spherical", although it can be inferred under most circumstances.
examples:
output = topojson.simplify(topology,{"minimum-area": 2,"coordinate-system": "spherical"});
output =topojson.simplify(topology,{"retain-proportion: 2,"coordinate-system": "spherical"});
I am not really sure exactly what the values you pass into these options mean, however higher values tends to produce more simplification. As a note, retain proportion often returns invalid topologies when passed LineStrings, that may be as intended.
Additionally using the quantization option in topojson.topology can be used to create a smaller, simpler output and may be the best solution to some similar use cases and also doesn't have any clearly documented server API examples anywhere so:
//very simplified, small output
topojson.topology({routes: routesCollection},{"quantization":100});
//very unfiltered, large output
topojson.topology({routes: routesCollection},{"quantization":1e8});
note: the default quantization is 10000 (1e4), so anything less than 10000 will create a smaller output and vice versa.
I'm looking for a way to specify that the images returned by the Google Custom Search API have a square format.
I've tried tbs=iar:s (because I've read Using the Custom Search API (REST JSON) to search for square images), but it doesn't work.
Have you an idea please ?
The problem is that tbs query parameter only applies to a regular image search on Google. For example, if you wanted to search for cat pictures with a square aspect ratio, you could do a search like this:
http://images.google.com/?q=cat&tbs=iar:s
But the Custom Search API uses a completely different set of parameters. The full list of supported parameters is shown in the REST documentation.
Some of the tbs queries do have equivalents. For example:
tbs=ic:gray translates to imgColorType=gray
tbs=isz:m translates to imgSize=medium
tbs=itp:clipart translates to imgType=clipart
But sadly there appears to be no equivalent for the iar aspect ratio filter. I tried guessing a few queries (things like imgAspectRatio=square) in case there was an undocumented parameter, but didn't have any luck with that.
The best alternative I could suggest is using imgSize=icon. This tends to return images that have a square aspect ratio, but with the unfortunate side effect that the images also tend to be rather small (the largest size I've seen returned is 256x256). Depending on your needs though, this may be good enough.
I apologise if this isn't particularly useful to you. I'm not just trying to grab the bounty on this question, so feel free not to vote this answer up. I just wanted to let you know what I had found in case it was of some help.
You can simply use both tbs=isz:l,iar:s that way it will return only large images with same aspect ration.