Python Bridge

Python is an popular interactive script language, created by Guido van Rossum and first released in 1991. Spaces, tabs and line feeds matter in Python code, and enforce a clean code style with correct indentations. Python is interpreted and thus slow compared to compiled languages like C. So it is not suited for high-speed strategies. But its strengths are a simple and modern syntax, and access to all sorts of libraries for machine learning and data analysis. Many tasks can be solved in the same way in lite-C as well as in Python, but Python often has already a dedicated command implemented or library available for that.

There are two methods to run Python code from Zorro. The Python bridge (Zorro 2.04 and above) allows directly calling Python functions from a Zorro script and using their results for trade signals. A script can start a Python session, send price data or indicator values to it, and use Python machine learning packages for training and prediction. Alternatively, Zorro can directly call the Python interpreter pythonw.exe and exchange data via files. This method is slower, but works even when the used Python library does not support embedded Python (such as NumPy). The examples below have code for both methods.

Installation and test

• Install Python 3.6.8 32 bit for Windows from https://www.python.org (even if you already have a newer Python version).
• Open Zorro.ini and edit the entry PythonPath to your Python folder
  (f.i. C:\Users\You\AppData\Local\Programs\Python\Python36-32).
• Start Zorro and run the script PythonTest. If everything went well, it prints out 0, 10, 20, 30, 40 and the sum 100.
• Move to the Python 3.6.8 folder and install the needed packages with pip: python -m pip install [packagename]
• For updating an existing package to the latest version, use python -m pip install [packagename] --upgrade

Make sure that any Python package is installed to your Python 3.6.8 folder, so don't use the pip --user option. Afterwards the package can be accessed with the import statement. Please read the remarks about issues with particular packages.

The following functions are available for sending data to Python, performing computations, and receiving data back:

pyStart(string Filename, int Mode): int

Parameters:

pyX(string Code): int

Parameters:

pySet(string Name, int n)

pySet(string Name, var d)

pySet(string Name, var *v, int elements)

Parameters:

pyInt(string Name): int

pyVar(string Name): var

pyVec(string Name, var *v, int elements)

Parameters:

Remarks:

• Numpy has issues with embedded Python. This also affects all libraries that use numpy, f.i. scipy, pandas, keras, tensorflow. Although numpy release 1.16.3 or above does start, it does not shut down properly, so you can only run it once with the Python bridge. Afterwards Zorro must be closed and restarted. Numpy was also reported to be not compatible with the R bridge, so r.h and contract.c cannot be included in numpy scripts. For overcoming all those issues, do not run numpy functions directly, but call the Python interpreter as in the example below.
• Test your Python commands and functions carefully with the Python console before executing them in a lite-C script. A command error - wrong syntax, a wrong path, a missing Python package, or a bad parameter for a Python function - can cause the session to abort and subsequent Python commands to fail.
• If you have the choice, perform script-based computations on the lite-C side. Lite-C is about 100 times faster than Python. 
• You can use [Train] mode and WFO for training Python machine learning algorithms, f.i. a SVM or neural network. For this, send training data to Python at the end of every WFO cycle, and store the trained models in files for later use in the [Test] or [Trade] session, similar to the neural function for R. A neural function for Python will be provided in a future Zorro version.

Examples:

// run some Python code
function main()
{
  if(!pyStart(0,1)) {
    printf("Error - Python won't start!");
    return;
  }

  var Vec[5] = { 0,1,2,3,4 };
  pySet("PyVec",Vec,5);
  pyX("for i in range(5): PyVec[i] *= 10\n");
  pyVec("PyVec",Vec,5);

  int i;
  printf("\nReturned: ");
  for(i=0; i<5; i++) printf("%.0f ",Vec[i]);

// test a function
  pyX("def PySum(V):\n Sum = 0.0\n for X in V:\n Sum += X\n return Sum\n\n");
  pyX("Result = PySum(PyVec)");
  printf("\nSum: %.0f",pyVar("Result"));
}

// run a function from a Python script
function main()
{
  if(!pyStart("MyPythonScript.py",1)) {
    printf("Error - Python script won't start!");
    return;
  }
  pyX("Result = MyPythonFunction()");
  var Result = pyVar("Result");
  printf("\nResult: %.2f",Result);
}

// start Python interpreter and process data

// Python side: --------------------------------
#read signals from file
Signals = []
with open("Data/In.txt","r") as File:
  for Line in File:
    Signals.append(float(Line))
# process signals and write result back to file
Processed = processMyData(Signals)
with open("Data/Out.txt","w") as File:
  for Line in Processed:
    File.write(str(Line))
    File.write('\n')


// Zorro side: --------------------------------
/ send signals to file 
string Content = zalloc(NumSignals*32);
for(i=0; i<NumSignals; i++)
  strcat(Content,strf("%.4f\n",Signals[i]));
file_write("Data\\In.txt",Content,0);
// call the Python interpreter 
string PythonPath = "MyPythonPath\\Pythonw";
exec(PythonPath,"Strategy/MyPython.py",1);
// get returned signals
file_read("Data\\Out.txt",Content,NumSignals*32);
Signals[0] = atof(strtok(Content,"\n"));
for(i=1; i<NumSignals; i++)
  Signals[i] = atof(strtok(0,"\n"));
...

See also: