logP training fails

User 44aac1fba7

09-02-2010 10:09:33

 My attempts at training logP fail in strange ways: I got 'logPparameters.txt' with several NaN values, when trying this under Windows XP (Calculator v5.3.0.2):


cxcalc -T logP -t train_LogP -o logPparameters.txt PlogP.sdf


 and later on a PluginException is thrown when trying the user-defined logP method.


When doing the same thing under Linux (Ubuntu 2.6.31-18-generic), cxcalc simply stops quietly without either an output or error messages produced!


 Please help me with getting this feature work.


-- Zoli

ChemAxon e08c317633

09-02-2010 11:00:06

Hi Zoli,


Have you copied the generated logPparameters.txt file to the Marvin config directory, as described in General remarks about the logP teaching?


What is the exception you get?


 


We will implement a graphical user interface for logP training. It will be available in Marvin 5.4.


Zsolt

User 44aac1fba7

10-02-2010 14:08:51

 Well, the error I got was:


1       logp: chemaxon.marvin.plugin.PluginException: No dataset found for user
defined method.
id      logP
1
...


 Since I'm using JChem, I had copied the data file into:


ChemAxon\JChem\marvin\config\logPparameters.txt


 But, upon re-reading the online instructions, this may not be the right place for it.


Should it go under


ChemAxon\JChem\MarvinBeans\ ?


This is for the Windows version, in which my cxcalc, version 5.3.0.2, is located at:
C:\UnixOid\Chem\ChemAxon\JChem\bin\cxcalc.bat


In any event, the Linux version produces no output whatsoever - so there's nothing to copy from... (Alas, I have nothing in the install tree like ChemAxon/MarvinBeans as suggested by <http://www.chemaxon.com/marvin/help/calculations/calc_training.html#logptraining>;; rather, the install directory is ChemAxon/JChem/Marvinbeans!)


 


-- Zoli, un-enlightened

ChemAxon e08c317633

11-02-2010 10:26:31










zfekete wrote:

This is for the Windows version, in which my cxcalc, version 5.3.0.2, is located at:
C:\UnixOid\Chem\ChemAxon\JChem\bin\cxcalc.bat



If this is the place where cxcalc is located, then the logPparameters.txt file should be copied to C:\UnixOid\Chem\ChemAxon\JChem\config directory.


Zsolt

User 44aac1fba7

12-02-2010 11:24:52

 OK, this way (which was kind of impossible to figure out, from the instructions specifically saying marvin/config, though) the training finally works under Windows.


 What about the Linux version stopping cold in its track?














 


Zsolt wrote:












zfekete wrote:

This is for the Windows version, in which my cxcalc, version 5.3.0.2, is located at:
C:\UnixOid\Chem\ChemAxon\JChem\bin\cxcalc.bat



If this is the place where cxcalc is located, then the logPparameters.txt file should be copied to C:\UnixOid\Chem\ChemAxon\JChem\config directory.


Zsolt


ChemAxon e08c317633

12-02-2010 11:55:01

The logPparameters.txt file should be copied to


- if MarvinBeans is installed: <path to MarvinBeans>/MarvinBeans/config


- if JChem is installed: <path to JChem>/JChem/config


directory.


If the default installation is followed, then on window OS  <path to MarvinBeans> = <path to JChem> = "C:\Program Files\ChemAxon".


Some help for linux. Execute "which cxcalc" from command line.


$ which cxcalc
<some path>/bin/cxcalc


The result is path, which shows the path to cxcalc. Copy the logPparameters.txt file here:


<some path>/config/


Where <some path> is the path to your MarvinBeans or JChem installation directory.


I hope this helps.


Zsolt

User 44aac1fba7

12-02-2010 13:55:55

> [...] Copy the logPparameters.txt file here:


 Well, like I said in the first round:


 *** THERE IS NO logPparameters.txt GENERATED *** -


the training run under Linux just stops without any output whatsoever!


-- Zoli

ChemAxon e08c317633

12-02-2010 14:12:43

OK, sorry. We will check this error.


Please run the "java -version" command on your linux system, and post the output here. It should look like this:


$ java -version
java version "1.6.0_16"
Java(TM) SE Runtime Environment (build 1.6.0_16-b01)
Java HotSpot(TM) Client VM (build 14.2-b01, mixed mode, sharing)


Zsolt

User 44aac1fba7

12-02-2010 14:20:56

java version "1.6.0_0"
OpenJDK Runtime Environment (IcedTea6 1.6.1) (6b16-1.6.1-3ubuntu1)
OpenJDK 64-Bit Server VM (build 14.0-b16, mixed mode)

ChemAxon e08c317633

12-02-2010 14:42:44

Please install JRE/JDK distributed by Sun. OpenJDK does not work always as expected ...


For more see System Requirements.


Zsolt

User 44aac1fba7

16-02-2010 08:16:36

 Oh, the wonderful world of almost-compatible Javas ;-<.


Now I reinstalled Sun's JRE+JDK, and uninstalled Ubuntu's default openjdk (breaking countless other applications, of course). 'java -version' says:



java version "1.6.0_15"
Java(TM) SE Runtime Environment (build 1.6.0_15-b03)
Java HotSpot(TM) Client VM (build 14.1-b02, mixed mode, sharing)


 And I'm back to square 1: 'cxcalc -T logP -t train_LogP -o logPparameters.txt PlogP.sdf' gives no output whatsoever!


 Strangely, while cxcalc was churning away to produce nothing, I noticed that an intermittent 'logPparameters.txt' was created (along with the aptly named 'naugye.txt') - this was erased automagically when the run completed, however!!!


 Ouch.

ChemAxon e08c317633

17-02-2010 10:09:38

Could you attach the PlogP.sdf file? If it contains confidential data, then you can contact us in email.


Zsolt

User 44aac1fba7

19-02-2010 07:43:22

 I emailed to support@chemaxon.com the PlogP.sdf input file.


Meanwhile I also run the test with Chemaxon's own CXtrainingset.sdf example file -

this time the run aborted with a message as follows:



Exception in thread "main" java.lang.NullPointerException

        at sun.misc.FloatingDecimal.readJavaFormatString(Unknown Source)

        at java.lang.Double.parseDouble(Unknown Source)

        at chemaxon.calculations.logPFragmentCreator.calculatelogPFragments(Unknown Source)

        at chemaxon.marvin.Calculator.runTrainingDataGenerator(Unknown Source)

        at chemaxon.marvin.Calculator.main(Unknown Source)


 

ChemAxon e08c317633

19-02-2010 10:42:49

I think you run the command


cxcalc -T logP -t train_LogP -o logPparameters.txt trainingset.sdf


There is no  "train_LogP" SDF field in our trainingset example (trainingset.sdf), that's why you get a NullPointerException. We will replace the error message with a more informative one.


Please run the


cxcalc -T logP -t LOGP -o logPparameters.txt trainingset.sdf


or


cxcalc -T logP -o logPparameters.txt trainingset.sdf


command, these should work.


We received the PlogP.sdf file, but we need some time to figure out why it doesn't work on Ubuntu linux (it works on Windows).


Zsolt

User 44aac1fba7

19-02-2010 11:08:18

 In the test with the Chemaxon example input, putting the erronous filename with '-t train_LogP' was my mistake, indeed - provoking the exception reported. Correcting the filename, then, gives the very same non-result (no output produced - or rather intermediately made one erased at the end) as my own training SDF, however!


 It does indeed works under Windows, as I mentioned previously. As a matter of fact, the Windows-produced logPparameters.txt can even be used under Linux, subsequently!

ChemAxon e08c317633

19-02-2010 16:09:55

We found the bug. The output file is not generated on Linux only if the file path after the "-o" option is "logPparameters.txt". So you can either specify a different path, or specify a different name for the file, and after it's generated rename it.


1.


mkdir temp
cxcalc -T logP -t LOGP -o temp/logPparameters.txt trainingset.sdf


2.


cxcalc -T logP -t LOGP -o mylogPparameters.txt trainingset.sdf
mv mylogPparameters.txt logPparameters.txt


In next version we will fix this bug, so generating files named logPparameters.txt will be possible.


Sorry for the inconvenience.


Regards,


Zsolt