Skip to content

Problem in conversion of regulate time series and forecasting using Date Time [Timestamp values]:R

4 messages · Dhivya Narayanasamy, Jim Lemon, Jeff Newmiller

Original
Dhivya Narayanasamy
# 
Hi,
I am new to R. Kindly help me with the plot that gives wrong x-axis
values.  I have a data frame "gg", that looks like this:
timestamps      value
1 2017-04-25 16:52:00 -0.4120000
2 2017-04-25 16:53:00 -0.4526667
3 2017-04-25 16:54:00 -0.4586667
4 2017-04-25 16:55:00 -0.4606667
5 2017-04-25 16:56:00 -0.5053333
6 2017-04-25 16:57:00 -0.5066667

I need to plot this as a Time series data to do forecasting. The steps are
as follows:

1) gg$timestamps <- as.POSIXct(gg$timestamps, format = "%Y-%m-%d %H-%M-%S")
 #changing "Timestamps" column 'factor' to 'as.POSIXct'.

2) gg.ts <- xts(x=gg$value, order.by = gg$timestamps) #converting the
dataframe to time series (Non Regular Time series)

3) fitting <- auto.arima(gg.ts) #fitting the time series model using
auto.arima

4) fore <- forecast(fitting, h=30, level = c(80,95))  #Forecasting

5) I am using plotly to this forecast model (Inspired from here :
https://plot.ly/r/graphing-multiple-chart-types/#plotting-forecast-objects)

plot_ly() %>%
  add_lines(x = time(gg.ts), y = gg.ts,
            color = I("black"), name = "observed") %>%
  add_ribbons(x = time(fore$mean), ymin = fore$lower[, 2], ymax =
fore$upper[, 2],
              color = I("gray95"), name = "95% confidence") %>%
  add_ribbons(x = time(fore$mean), ymin = fore$lower[, 1], ymax =
fore$upper[, 1],
              color = I("gray80"), name = "80% confidence") %>%
  add_lines(x = time(fore$mean), y = fore$mean, color = I("blue"), name =
"prediction")


The plot comes out wrong: 1) x axis labels are wrong. It shows some
irrelevant values on axis. 2) the plot is not coming out.
Also I tried to convert "gg.ts" to a regulate time series which throws
error :
Error in 1/frequency : non-numeric argument to binary operator

Please help me how to use Date Time values in converting to regulate time
series for forecasting.


Regards

  
  


  


    

    

      
      

        


  

        

      Jim Lemon
    

    

      
      #
    

  


  

      
Hi Dhivya,
I'm not that familiar with the "gg.ts" function, but you are passing
character values to the "frequency" and "start" arguments. If there is
no automatic conversion to numeric values, that would cause the error.
Similarly, your "timestamps" variable may have been read in as a
factor, which often causes trouble with date conversions. Try
as.character(gg$timestamps) instead of just gg$timestamps.

Jim


On Thu, Apr 27, 2017 at 4:51 PM, Dhivya Narayanasamy

        
<dhiv.shreya at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Dhivya Narayanasamy
    

    

      
      #
    

  


  

      
Hi Jim,

Thank you for the reply. 'gg.ts' is actually the object name of the time
series I am using here.  Also I have changed my timestamp class from factor
to POSIXct  (gg$timestamps <- as.POSIXct(gg$timestamps, format = "%Y-%m-%d
%H-%M-%S") . When i plot this time series on graph, the x axis scales shows
random value rather than showing timestamp value. Is there any way to
correct the graph to make it show the timestamp value on x axis?  I use
plotly function for plotting.

Regards| Mit freundlichen Gr??en,

Dhivya Narayanasamy

Contact No: +91-8438505020

        
On Fri, Apr 28, 2017 at 3:19 AM, Jim Lemon <drjimlemon at gmail.com> wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Jeff Newmiller
    

    

      
      #
    

  


  

      
gg$timestamps <- as.POSIXct(as.character( gg$timestamps ) )

Factors are integers with the appearance of character data, so you are converting the integers to POSIXct. I usually try to avoid letting R automatically convert character data to factors, e.g. using stringsAsFactors=FALSE in read.csv so I don't have to convert them back from factors to manipulate them.