TE in complex systems close to transition

[AmirHossein Hassanzadeh]

Hello,

I've been recently trying to reproduce the results of the following article with JIDT:

https://lib.ugent.be/fulltxt/RUG01/002/508/648/RUG01-002508648_2018_0001_AC.pdf

But unfortunately I'm receiving very different results as I'm feeding the time series to the JIDT to calculate TE. I would really be thankful if anyone lets me know what I am doing wrong. 

On page 24 you can see some graphs which are plotted using the author's code. I used the time series and the default settings of "k_HISTORY" , "k_TAU" , "I_HISTORY" , "I_TAU" and noise "NOISE_LEVEL_TO_ADD"  and this was the result.

The code I used was like this:

def javCalcPTE (dataRaw , src , dest):

    

    # Add JIDT jar library to the path

    jarLocation = "D:\AmirHossein\jlizier-jidt-64a7a80\infodynamics.jar"

    # Start the JVM (add the "-Xmx" option with say 1024M if you get crashes due to not enough memory space)

    

    

    if(not isJVMStarted()):

        startJVM(getDefaultJVMPath(), "-ea", "-Djava.class.path=" + jarLocation)

    # As numpy array:

    data = np.array(dataRaw)

    

    # 1. Construct the calculator:

    calcClass = JPackage("infodynamics.measures.continuous.kraskov").TransferEntropyCalculatorKraskov

   

    calc = calcClass()

    

    # 2. Set any properties to non-default values:

    # No properties were set to non-default values

    source = data[:,src]

    destination = data[:,dest]

    # 3. Initialise the calculator for (re-)use:

    calc.initialise()

    # 4. Supply the sample data:

    calc.setObservations(source, destination)

    

    # 5. Compute the estimate:

    result = calc.computeAverageLocalOfObservations()

    

    return result

Also as someone recommended I used the matrix for of TE and the difference of the target time series as target, i.e. 

# Compute for all pairs:

    for s in range(3):

        for d in range(3):

            # For each source-dest pair:

            if (s == d):

                continue

            source = data[:-1 , s]

            destination = np.diff(data[: , d])

            # 3. Initialise the calculator for (re-)use:

            calc.initialise()

            # 4. Supply the sample data:

            calc.setObservations(source, destination)

            # 5. Compute the estimate:

            PTE[s][d] = calc.computeAverageLocalOfObservations()

            #print (PTE[s][d])

    #print (PTE)

    return PTE

And this was the result:

These were the results I was comparing with the TE plot (4th plot on the left).

Thank you, 

AmirHossein Hassanzadeh

P.S: Using the code on the thesis, I gained the same results as shown in page 24 of the thesis.