Excellent presenter. Very clearly and well explained. The video editing is excellently done. I prefer just to see the slides, but if you're going to show the presenter this is exactly how you should do it; you see both the presenter and the slides. Thank you for a job well done.
Hi Brett , Can I use that great tool to dabble with a multi-varte time series with 4 columns + timestamp column where such 4 columns do interact with each other , it is called order book , and can we render online version of such classification time series algos , in order to save computing and storing resources ? Your input is highly appreciated
Did he say that data time series data is collected at irregular time periods? Doesn't this violate the definition of a time series? I would classify that type of data as longitudinal rather than time series...
Kinda difficult to comment without a clear scenario, so I'll take a univariate series ARMA(p,q) as an example with LOESS as the interpolation method. If your data is uniformly "missing" or "lacking" values then LOESS (with appropriate tunning) should be robust enough to handle that. However, if your series appears more like an an ARIMA(p,d,q) series (with nonstationary in mean and variance) and your data is lacking in observations non-uniformly, then this creates issues for LOESS as it depends on the dense (as in many) local observations to estimate what values should fall in. LOESS can then over/under estimate regardless of your tunning. Another situation is that if LOESS is used, do you then not used the original interval points for which the original observations are measured, or do you move to the interval values by which your interpolation method produced values? This gets more complicated with multivariate time series of course. There are other topics in density estimation that will also overlap. These are just some of my thoughts.
Thankfully the music was just 30 secs...
26 seconds...
Excellent presenter. Very clearly and well explained. The video editing is excellently done. I prefer just to see the slides, but if you're going to show the presenter this is exactly how you should do it; you see both the presenter and the slides. Thank you for a job well done.
Thanks for posting. This is an area of interest for me. Looking for more materials and peers on time series analysis.
Hi Brett ,
Can I use that great tool to dabble with a multi-varte time series with 4 columns + timestamp column where such 4 columns do interact with each other , it is called order book , and can we render online version of such classification time series algos , in order to save computing and storing resources ? Your input is highly appreciated
This is really helpful! Thank you :')
See the complete SciPy 2016 Conference talk & tutorial playlist here: th-cam.com/play/PLYx7XA2nY5Gf37zYZMw6OqGFRPjB1jCy6.html
nice music in the beginning :p
Why does everyone hate fullscreen browsers? :-)
Did he say that data time series data is collected at irregular time periods? Doesn't this violate the definition of a time series? I would classify that type of data as longitudinal rather than time series...
Interpolate?
That's what would be required before the analysis. However, interpolation creates another dimension of obstacles to account for.
Can you explain further? I understand generally that interpolation introduces issues but I'd like to hear your thoughts.
Kinda difficult to comment without a clear scenario, so I'll take a univariate series ARMA(p,q) as an example with LOESS as the interpolation method. If your data is uniformly "missing" or "lacking" values then LOESS (with appropriate tunning) should be robust enough to handle that. However, if your series appears more like an an ARIMA(p,d,q) series (with nonstationary in mean and variance) and your data is lacking in observations non-uniformly, then this creates issues for LOESS as it depends on the dense (as in many) local observations to estimate what values should fall in. LOESS can then over/under estimate regardless of your tunning. Another situation is that if LOESS is used, do you then not used the original interval points for which the original observations are measured, or do you move to the interval values by which your interpolation method produced values? This gets more complicated with multivariate time series of course. There are other topics in density estimation that will also overlap. These are just some of my thoughts.
Jonathan Navarrete he says that some time series are irregularly spaced...which does not violate the definition of a time series