mps9506 / mann-kendall-trend Goto Github PK

Compared to http://vsp.pnnl.gov/help/Vsample/Design_Trend_Mann_Kendall.htm, the code is adding the term for tied values instead of substracting it.
I can do a pull-request if necessary.

could be implemented with a numpy trick:

s = np.sum(np.sign(np.tril(np.subtract.outer(a,a)))) # sum of sign(xi-xj) where i>j

Hi,

When running the MK function I am getting an error message on some datasets:
test_trend,h,p,z = mk_test(ts,alpha=0.05)
File "C:/Python/Anaconda/SA_Water/TrendTest_rainfall_seasonal_data.py", line 107, in mk_test
tp[i] = sum(unique_x[i] == x)
File "C:\Python\Anaconda\lib\site-packages\pandas\core\ops.py", line 741, in wrapper
if len(self) != len(other):
TypeError: len() of unsized object

Any ideas what this is about?

When it runs it is brilliant!!
Regards,

Darren

Hi,

Are there any plans to implement the algorithm to calculate the number of samples required to detect a trend? As per the relevant section in http://vsp.pnnl.gov/help/Vsample/Design_Trend_Mann_Kendall.htm.

If not, then I'd be happy to attempt a PR.

Thanks.

If null hypothesis = data comes from population with independent realizations and are identically distributed, the alternative hypothesis is that the data follows a monotonic trend. This is calculated with the two tailed test and currently implemented in the script.

However, if the alternative is indeed an "increasing trend" or "decreasing trend" (as currently returned in the script) p values should be calculated with the single tail of the distribution.

See:
Implementation in the trend package in R.

Suggest adding an argument for users to specify the alternative.

Thanks to Moshe Elitzur for identifying this.

Add BSD license to make compatible with scipy: https://github.com/scipy/scipy/blob/master/LICENSE.txt