search

Teichlab / cellphonedb

MIT License
342 stars 105 forks source link

P-value Format #349

Closed julie-jch closed 2 years ago

julie-jch commented 3 years ago

Hi team, I was working with your results file and noted that p-values are rounded to 1 decimal place. Is there any way the actual p-values without any rounding can be obtained? Thanks in advance.

prete commented 2 years ago

Hi @julie-jch thank you for using CellPhoneDB. The rounding is handled by the --result-precision option when launching CellphoneDB, by default that value is 3. You probably only see 1 decimal place for results in pvalues.txt with a value of 1.0, are you seen all values with only 1 decimal?

Nicolas-Eng commented 2 years ago

hi @prete, i changed my --result-precision to 10, but it seems to have changed the mean values instead of the p-values. Is there any fix for this?

julie-jch commented 2 years ago

Thanks @prete I changed the result precision and increased my iterations and was able to get 2dps.

Nicolas-Eng commented 2 years ago

@julie-jch how many iterations did you use?

julie-jch commented 2 years ago

@Nicolas-Eng I first ran it using 1000 and now I'm currently running it at 10000, hoping to get a better result.

Nicolas-Eng commented 2 years ago

@julie-jch thank you, I will run mine at 10000 with a precision of 10 and let you know my results as well.

julie-jch commented 2 years ago

So running with the following parameters: 

          _iterations=10000
          threads=10 
          result-precision 100_

got me 3dps (which is good enough for my analysis). I launched the analysis at 10:32:37 and the results populated at 15:10:39, so it took 4hr 38 minutes to run.