It would have been great if you could have used an external microphone rather than the internal laptop microphone. As your PC is getting heated, the noise of its fans is acting as a major 'irregular factor', and disrupting the voice.
+orslo Thanks for that key insight!!! I will definitely put this into play and appreciate your comment very much. Thank you. I just bought an external microphone and am going to try and fix the audio issues.
Thanks for the number of examples you explain! Ive got two questions: At 1:09:00 you choose arima over holtwinters because de irreguler component had values of 3500 and 4500. Why? And when you use this arima to forecast the next few time steps, it predicts the seasonality nicely. How did the arima model find the seasonality?
Sorry for asking multiple questions but you mentioned something I thought was very powerful which was to subtract the seasonal component from the original time series so you're left with just Trend + Random. I strongly want to know the "how" in how you plotted the Seasonal plot and how I can do the same. One approach I've taken in the past is to take the mean of say 12 months and subtract the average of each month from the mean to get an idea of percentage each month is either above or below the mean. Any tips? My language of choice is Python but any input you could suggest would be very helpful and carry much more weight than the questions I read on blogs or stackoverflow.
+jnscollier I am pretty sure that we can use the Rpy2 library to invoke R within Python. This might be helpful because then you can draw from pandas, numpy, scipy, and also tap into R within the same framework. As to your question, in R there is a package called "forecast" which was developed by a pioneer in the time series space, Professor Rob Hyndman. What I will typically do here with his package is take the dataframe, convert it into a ts object, decompose the components, and then strip out just the seasonally adjusted portion. Here some sample R code for for monthly data starting in 1-2010: 1. libarary("forecast") 2. mydata
I'm very confused here: While explaining Holt's Exponential smoothing around 37 minutes in the video, you keep on doing statistical tests for 'Simple Exponential smoothing'. And later at 38:33 you conclude that from the results of the statistical tests, you we can conclude that the 'Holt's exponential' smoothing provides adequate model that can't be improved upon. Please clarify!
Hi! Can you recommend any other sources about 13:20? Where you just choose an appropriate algorithm based on trend, seasonality, and correlation.. I found many sources telling me how to do these methods, but I found non that would tell me which one should I use given a particular data. Please reply asap. I really need you help. Thank you. :)
+Mohammed Shabeeb Thanks for the kind words. I am definitely thinking about going back and cleaning up the audio content. I originally recorded this directly through ppt and a microphone on my surface. Hopefully there is an easy way to get this clean and crisp that wont involve me having to re-record everything. In the meantime, I hope that it is not a deal breaker for you. Good luck and thanks for watching.
i have research about analysis time series to eye movement i extraction feature speed eye,velocity,dirction (tan),and acceleration ,i have dirction (tan) and time can i use holts exponential smoothing to make forecasts? can you give me another choices about that thank u
If ACF and PACF tail off at lag 10 in a lag=40 plot but has a few lag points that also breach the shaded region in the correlograms later on, does this mean ARIMA might not be a good model? www.dropbox.com/s/nirz3nh9ydiahkr/correlogram.png?dl=0
+jnscollier Not necessarily, but probably... :0) There will probably be candidate ARMA models with with (p,q) of (2,0) based on the ACF, and (10,0) based upon the PACF. The principle of parsimony would state that the ARMA(2,0) would be preferred because it is simplier. However, I wonder how effective the model will be with the such high # of lags.
+Lee jacquie Thanks. Please pm me and I will get you access to the dropbox with the code. I will eventually set up a github account but hopefully this will work in the meantime. Thanks for watching and good luck.