Recipies and FAQ for the Timeseries Scikit
- General threads
- Reading data and creating timeseries objects
- About this page
NOTE: The official documentation and important remarks from the developers can be found at the timseries scikit sourceforge page.
time series analysis - http://article.gmane.org/gmane.comp.python.scientific.user/13949
- time series: Python vs. R URL missing!!!
roadmap/plans for timeseries package - http://permalink.gmane.org/gmane.comp.python.scientific.user/14599
Reading data and creating timeseries objects
masking NoData values
In my original data nodata values are marked with "-999". How can I import the data or create the time series and exclude these no data points from further processing? (flagging no data in timeseries - http://permalink.gmane.org/gmane.comp.python.scientific.user/14455)
- use masked_where from maskedarray
1 myvalues_ts_hourly = masked_where(myvalues_ts_hourly , -999)
- Use indexing
1 myvalues_ts_hourly[myvalues_ts_hourly==-999] = M.masked
More extensive answer
* START SAMPLE DATA (tmp.txt) *
* END SAMPLE DATA *
1 import numpy as N 2 import maskedarray as M 3 import timeseries as ts 4 data = N.loadtxt("tmp.txt", dtype='|S10', skiprows=2) 5 dates = ts.date_array([ts.Date(freq='H',string="%s %s:00" % 6 (d,int(d)-1)) 7 for d in data], 8 freq='H') 9 series = ts.time_series(data[:,-1].astype(N.float_), 10 dates, 11 mask=(data[:,-1]=='-999'))
Is there a example data set for at least one year on a high temporal resolution: 15min or at least 1h. Having such a common data set one could set up tutorials examples and debug or ask questions easier because all will have the same (non-confidetial) data on the disk.
For hours, you have the 'hourly' frequency. For 15min, you have the 'minutely' frequency, from which you can select every other 15th point.
(cf. Re: roadmap/plans for timeseries package - http://permalink.gmane.org/gmane.comp.python.scientific.user/1459)
hour of the day
(cf.: assignment of hours of day in time series - http://permalink.gmane.org/gmane.comp.python.scientific.user/14597) When exchanging agrregated data sets (e.g. with hourly frequency) the data is often presented as follows: desired report output
This formatting may be the result of some logging devices which for instance record 5 minutes averaged values which have been taken with a device using a sample rate of 16 sec. As well, syntetically generated data sets which have been created by scientifc models or from remote sensing information can have such a format. When creating a timeseries object the start hour should be set to zero (0) internally to achieve a correct assignment of the hours (01:00 h is the end of the period 00:00 h - 01:00 h => data for this period starts at 00:00 h). For the output one can be customized as shown below in the answer. The python built-in module datetime can help here.
I have hourly measurements where hour 1 represents the end of the period 0:00-1:00, 2 the end of the period 1:00-2:00, ... , 24 the end of the period 23:00 to 24:00.
When I plot these hourly time series from February to November the curve is continued into December because of that matter. time series then assumes that the value for hour 0:00 of dec, 01 is 0 which then leads do a wrong plotting behaviour.
I want to achieve that hour 24 is accounted as the last measurement period of a day and not as the first measurement of the next day (like 0:00).
Since the time "24:00" doesn't actually exist (as far as I am aware anyway), you will have to rely on somewhat of a hack to get your desired output. Try this:
Manipulations & Operations with time series
use the datetime information of the time series
(Re: roadmap/plans for timeseries package - http://permalink.gmane.org/gmane.comp.python.scientific.user/14598) A example:
One has to get rainfall intensity during early morning hours. For such a filter the information on the corresponding hours are neccessary.
1 import timeseries as ts 2 data = ts.time_series(range(100), start_date=ts.today('hourly')) 3 hours = data.hour 4 filtered_data = data[(hours < 7) & (hours > 3)] 5 filtered_data 6 timeseries([80 6 7 8 30 31 32 54 55 56 78 79], 7 dates = [07-Jan-2008 04:00 07-Jan-2008 05:00 07-Jan-2008 06:00 8 08-Jan-2008 04:00 08-Jan-2008 05:00 08-Jan-2008 06:00 09-Jan-2008 04:00 9 09-Jan-2008 05:00 09-Jan-2008 06:00 10-Jan-2008 04:00 10-Jan-2008 05:00 10 10-Jan-2008 06:00], 11 freq = H)
using the result of time series operations
How can one save the read the result of time series operations into a array?
For instance, if I convert data in an hourly frequency to daily averages how to I read the daily averages into a array for further processing?
when I print out my daily timeseries converted from hourly data I get something like this:
What I would like is an array with just the values of the daily averages . Additional a report-like array output with the format day value
1 3 2 11
> For instance, if I convert data in an hourly frequency to daily averages > > how to I read the daily averages into a array for further processing?
- possibility #1: use the keyword func while converting.
avgdata = convert(data,'D', func=maskedarray.mean)
- possibility #2:
If you don't use the keyword func, you end up with a 2d array, each row being a day, each column an hour. Just use maskedarray.mean on each row avgdata = convert(data,'D').mean(-1)
If you only want the values, use the .series attribute, it will give you a view of the array as a MaskedArray.
Word of caution... the timeseries plotting stuff does not currently support frequencies higher than daily (eg. hourly, minutely, etc...). Support for these frequencies could be added without too much trouble, but just haven't got around to it yet. (Cf. Re: roadmap/plans for timeseries package - http://permalink.gmane.org/gmane.comp.python.scientific.user/14598)
About this page
- Most information presented here has been compiled from discussions at the scipy mailing list.
- Use one data set consistently for the examples
- offer the code for download