nrnpython

neuron.h
cas

neuron.h.Section

neuron.hoc
execute · hoc_ac

neuron.hoc.HocObject

PythonObject

Warning

Some of the idioms on this page are out of date, but they still work. See the NEURON Python tutorial for modern idioms.

Python Language

This document describes installation and basic use of NEURON’s Python interface. For information on the modules in the neuron namespace, see:

Installation

Syntax:

./configure --with-nrnpython ...

make

make install

Description:

Builds NEURON with Python embedded as an alternative interpreter to HOC. The Python version used is that found from which python.

“make install” automatically does a (cd src/nrnpython ; python setup.py install --home=<prefix>) which generally puts the neuron module under <prefix>/lib/python. Note: This path then needs to be added to your PYTHONPATH environment variable. To install in the default system Python directory, manually do the cd and run python setup.py install (possibly with a sudo). Be advised that this installs only for one version of Python, so if you use both Python 2 and 3, keep reading.

Other python related configure arguments are:
--enable-pysetup=installoption
Execute ‘python setup.py install installoption’ as the last installation step. –disable-pysetup or an installoption of ‘no’ means do NOT execute ‘python setup.py…’ The default installoption is ‘–home=<prefix>’
--disable-rx3d Do not compile the cython translated 3-d rxd
features

--with-nrnpython-only configure and make only the nrnpython folder

--with-nrnpython=``desired python binary or ``dynamic
Python interpreter can be used (default is NO) Probably need to set PYLIBDIR to find libpython… and PYINCDIR to find Python.h

--with-pyexe``=desired python binary (when ``--with-nrnpython=dynamic)

--disable-pysetup skips the automatic invocation of python setup.py;
to use Python, this will have to be done manually.

I use this script to build a version of NEURON that works with python2.5-7 and python3.5-6

hines@hines-T7500:~/neuron$ cat bld-nrndynam.sh
#!/bin/bash

cd $HOME/neuron/nrndynam

../nrn/configure --prefix=`pwd` --with-paranrn=dynamic \
  --with-nrnpython=dynamic --with-pyexe=python3 --disable-rx3d
rm -r -f src/nrnpython/build
make -j 6 install

../nrn/configure --prefix=`pwd` --with-paranrn=dynamic \
  --with-nrnpython=dynamic --with-pyexe=python2 --disable-rx3d \
  --with-nrnpython-only
rm -r -f src/nrnpython/build
make -j 6 install

cd lib/python/neuron
cp hoc.cpython-35m-x86_64-linux-gnu.so hoc.cpython-36m-x86_64-linux-gnu.so
cd ../../..
# or alternatively configure ..; rm...; make... with other versions for --with-pyexe

Originally I thought that nrniv -python ... was the best way to launch (and perhaps is for parallel and on many supercomputers). Now, especially with the possibility of many python versions and one (–with-nrnpython=dynamic) version is it is simpler to launch the correct python with only the necessity of a uniform export PYTHONPATH=<prefix>/lib/python (and it is then a bad idea to install the neuron module in a specific python’s site-packages). This gives the possibility of python foo.py python2 foo.py python3.5 foo.py python3.6 foo.py Whereas launching nrniv would require an environment with the proper PYTHONHOME and PYNRNLIB The former to avoid the dreaded “site” problem and the latter to ensure the right python library is loaded. In vicious cases of failure to launch due to site problems, the first diagnostic info needed to fix the problem is

import site
print (site.__file__)

For launching nrniv,there is a script nrnpyenv.sh that helps determine a correct PYTHONHOME, etc for the default python executable.

If launching Python and using parallel neuron simulations there is the issue of who calls MPI_Initial. Default assumes Python will call it (perhaps via from mpi4py import MPI). If that is not the case, force the neuron module to initialize MPI when the neuron module is loaded by setting the environment variable NEURON_INIT_MPI

If you just can’t seem to get nrniv to launch and you don’t need Python, you can try nrniv -nopython ... or permanently add nopython: on to $HOME/.nrn.defaults or <prefix>/share/nrn/lib/nrn.defaults (c:/nrn/lib/nrn.def on windows)

You can use the hoc function nrn_load_dll() to load mechanism files as well, e.g. if neurondemo was used earlier so the shared object exists,

from neuron import h
h.nrn_load_dll("$(NEURONHOME)/demo/release/x86_64/.libs/libnrnmech.so")

Python Accessing HOC

Syntax:

nrniv -python [file.hoc file.py  -c "python_statement"]

nrngui -python ...

neurondemo -python ...

Description:
Launches NEURON with Python as the command line interpreter. File arguments with a .hoc suffix are interpreted using the Hoc interpreter. File arguments with the .py suffix are interpreted using the Python interpreter. The -c statement causes python to execute the statement. The import statements allow use of the following

Note

Most of the following is from the perspective of someone familiar with HOC; for a Python-based introduction to NEURON, see http://neuron.yale.edu/neuron/static/docs/neuronpython/index.html

class neuron.hoc.HocObject
Syntax:

from neuron import h

h = neuron.hoc.HocObject()

Description:

Allow access to anything in the Hoc interpreter. h is an instance of a neuron.hoc.HocObject object. Note that there is only one Hoc interpreter, no matter how many interface objects are created, so there is no advantage to creating another.

h("any hoc statement")

Any hoc variable or string in the Hoc world can be accessed in the Python world:

h('strdef s')
h('{x = 3  s = "hello"}')
print(h.x)          # prints 3.0
print(h.s)          # prints hello

And if it is assigned a value in the python world it will be that value in the Hoc world. (Note that any numeric python type becomes a double in Hoc.)

h.x = 25
h.s = 'goodbye'
h('print x, s')    #prints 25 goodbye

Note, however, that new Hoc variables cannot be defined from Python except via, e.g. h('strdef s').

Any hoc object can be handled in Python, and can use Python idioms for that type of object despite being created in hoc. e.g. in hoc, you would have to use vec.size() to get the Vector’s size. This still works in Python, but you can also use the Pythonic len(h.vec):

h('objref vec')
h('vec = new Vector(5)')
print(h.vec)        # prints Vector[0]
print(len(h.vec))   # prints 5.0

There is, however, in pure Python models never a need to create a hoc object; e.g. if no HOC code needed to access the Vector, the above is equivalent to

vec = h.Vector(5)
print(vec)
print(len(vec))

Note that any hoc object method or field may be called, or evaluated/assigned using the normal dot notation which is consistent between hoc and python. However, hoc object methods MUST have the parentheses or else the Python object is not the return value of the method but a method object. ie.

x = h.vec.size     # not 5 but a python callable object
print(x)            # prints: Vector[0].size()
print(x())          # prints 5

This is also true for indices

h.vec.indgen().add(10) # fills elements with 10, 11, ..., 14
print(h.vec[2])    # prints 12.0
x = h.vec.x        # a python indexable object
print(x)           # prints Vector[0].x[?]
print(x[2])        # prints 12.0

Note that the .x notation is not needed in Python for reading or (as of NEURON 7.7) writing to vectors.

The hoc object can be created directly in Python. E.g.

v = h.Vector(range(10, 20))

Iteration over hoc Vector, List, and arrays is supported. e.g.

v = h.Vector(range(10, 14))
for x in v:
  print(x)

l = h.List(); l.append(v); l.append(v); l.append(v)
for x in l:
  print(x)

h('objref o[2][3]')
for x in h.o:
  for y in x:
    print(x, y)

Any hoc Section can be handled in Python. E.g.

h('create soma, axon')
ax = h.axon

makes ax a Python Section which references the hoc axon section. Many hoc functions use the currently accessed section; most of these are now available as section methods, however for user written hoc and in legacy code, a “sec” keyword parameter temporarily makes the Section value the currently accessed section during the scope of the function call. e.g

print(h.secname(sec=ax))

Note

In Python, one can simply:

print(ax)

Or use str(ax) to get the name of the section ax.

Most such functions now have an alternative form that avoids the need for sec=; often they are available as section methods. This is usually listed in the function definition.

Point processes are handled by direct object creation as in

stim = IClamp(ax(1.0))

Many hoc functions use call by reference and return information by changing the value of an argument. These are called from the Python world by passing a HocObject.ref() object. Here is an example that changes a string.

h('proc chgstr() { $s1 = "goodbye" }')
s = h.ref('hello')
print(s[0])          # notice the index to dereference. prints hello
h.chgstr(s)
print(s[0])          # prints goodbye
h.sprint(s, 'value is %d', 2+2)
print(s[0])          # prints value is 4

and here is an example that changes a pointer to a double

h('proc chgval() { $&1 = $2 }')
x = h.ref(5)
print(x[0])          # prints 5.0
h.chgval(x, 1+1)
print(x[0])          # prints 2.0

Finally, here is an example that changes a objref arg.

h('proc chgobj() { $o1 = new List() }')
v = h.ref([1,2,3])  # references a Python object
print(v[0])          # prints [1, 2, 3]
h.chgobj(v)
print(v[0])          # prints List[0]

Unfortunately, the HocObject.ref() is not often useful since it is not really a pointer to a variable. For example consider

h('x = 1')
y = h.ref(h.x)
print(y)                     # prints hoc ref value 1
print('%g %g' % (h.x, y[0])) # prints 1.0 1.0
h.x = 2
print('%g %g' % (h.x, y[0])) # prints 2.0 1.0

and thus in not what is needed in the most common case of a hoc function holding a pointer to a variable such as Vector.record() or Vector.play(). For this one needs the _ref_varname idiom which works for any hoc variable and acts exactly like a c pointer. eg:

h('x = 1')
y = h._ref_x
print(y)                     # prints pointer to hoc value 1
print('%g %g' % (h.x, y[0])) # prints 1.0 1.0
h.x = 2
print('%g %g' % (h.x, y[0])) # prints 2.0 2.0
y[0] = 3
print('%g %g' % (h.x, y[0])) # prints 3.0 3.0

Of course, this works only for hoc variables, not python variables. For arrays, use all the index arguments and prefix the name with _ref_. The pointer will be to the location indexed and one may access any element beyond the location by giving one more non-negative index. No checking is done with regard to array bounds errors. e.g

v = h.Vector(range(10, 14))
y = v._ref_x[1]    # holds pointer to second element of v
print('%g %g' % (v[2], y[1])) # prints 12.0 12.0
y[1] = 50
v.printf()         # prints 10 11 50 13

The idiom is used to record from (or play into) voltage and mechanism variables. eg

from neuron import h
soma = h.Section(name='soma')
soma.insert('pas')
v = h.Vector().record(soma(0.5)._ref_v)
pi = h.Vector().record(soma(0.5).pas._ref_i)
ip = h.Vector().record(soma(0.5)._ref_i_pas)

The factory idiom is one way to create Hoc objects and use them in Python.

h('obfunc newvec() { return new Vector($1) }')
v = h.newvec(10).indgen().add(10)
v.printf()          # prints 10 11 ... 19 (not 10.0 ... since printf is a hoc function)

but that idiom is more or less obsolete as the same thing can be accomplished directly as shown a few fragments back. Also consider the minimalist

vt = h.Vector
v = vt(4).indgen().add(10)

Any Python object can be stored in a Hoc List. It is more efficient when navigating the List to use a python callable that avoids repeated lookup of a Hoc method symbol. Note that in the Hoc world a python object is of type PythonObject but python strings and scalars are translated back and forth as strdef and scalar doubles respectively.

h('obfunc newlist() { return new List() }')
my_list = h.newlist()
apnd = my_list.append
apnd([1,2,3])      # Python list in hoc List
apnd(('a', 'b', 'c')) # Python tuple in hoc List
apnd({'a':1, 'b':2, 'c':3}) # Python dictionary in hoc List
for item in my_list:
    print(item)

h('for i=0, List[0].count-1 print List[0].object(i)')

To see all the methods available for a hoc object, use, for example,

dir(h.Vector)

h.anyclass can be subclassed with

class MyVector(neuron.hclass(neuron.h.Vector)) :
  pass

v = MyVector(10)
v.zzz = 'hello' # a new attribute
print(v.size()) # call any base method

If you override a base method such as ‘size’ use

v.baseattr('size')()

to access the base method. Multiple inheritance involving hoc classes probably does not make sense. If you override the __init__ procedure when subclassing a Section, be sure to explicitly initialize the Section part of the instance with

nrn.Section.__init__()

Since nrn.Section is a standard Python class one can subclass it normally with

class MySection(neuron.nrn.Section):
  pass

The hoc setpointer statement is effected in Python as a function call with a syntax for POINT_PROCESS and SUFFIX (density)mechanisms respectively of

h.setpointer(_ref_hocvar, 'POINTER_name', point_proces_object)
h.setpointer(_ref_hocvar, 'POINTER_name', nrn.Mechanism_object)

See nrn/share/examples/nrniv/nmodl/(tstpnt1.py and tstpnt2.py) for examples of usage. For a density mechanism, the ‘POINTER_name’ cannot have the SUFFIX appended. For example if a mechanism with suffix foo has a POINTER bar and you want it to point to t use

h.setpointer(_ref_t, 'bar', sec(x).foo)

neuron.hoc.hoc_ac()
Syntax:

import hoc

double_value = hoc.hoc_ac()

hoc.hoc_ac(double_value)

Description:

Get and set the hoc global scalar, hoc_ac_-variables. This is obsolete since HocObject is far more general.

import hoc
hoc.hoc_ac(25)
hoc.execute('print hoc_ac_') # prints 25
hoc.execute('hoc_ac_ = 17')
print(hoc.hoc_ac())  # prints 17

neuron.h.cas()
Syntax:
sec = h.cas()
Description:

Returns the currently accessed section as a Python Section object.

from neuron import h
h('''
  create soma, dend[3], axon
  access dend[1]
''')

sec = h.cas()
print(sec)

It is generally best to avoid writing code that manipulatesd the section stack. Use Python section objects, sec=, and section methods instead.


class neuron.h.Section
Syntax:

sec = h.Section()

sec = h.Section([name='string', [cell=self])

Description:

The Python Section object allows modification and evaluation of the information associated with a NEURON Conceptual Overview of Sections. The typical way to get a reference to a Section in Python is with neuron.h.cas() or by using the hoc section name as in asec = h.dend[4]. The sec = Section() will create an anonymous Section with a hoc name constructed from “Section” and the Python reference address. Access to Section variables is through standard dot notation. The “anonymous” python section can be given a name with the named parameter and/or associated with a cell object using the named cell parameter. Note that a cell association is required if one anticipates using the gid2cell() method of ParallelContext.

import neuron
h = neuron.h
sec = h.Section()
print(sec)         # prints <nrn.Section object at 0x2a96982108>
print(sec.name())  # prints PySec_2a96982108
sec.nseg = 3       # section has 3 segments (compartments)
sec.insert("hh")   # all compartments have the hh mechanism
sec.L = 20         # Length of the entire section is 20 um.
for seg in sec:    # iterates over the section compartments
  for mech in seg: # iterates over the segment mechanisms
    print('%s %g %s' % (sec, seg.x, mech.name()))

A Python Section can be made the currently accessed section by using its push method. Be sure to use pop_section() when done with it to restore the previous currently accessed section. I.e, given the above fragment,

from neuron import h
h('''
objref p
p = new PythonObject()
{p.sec.push() psection() pop_section()}
''')
#or
print(sec)
h.psection(sec=sec)

When calling a hoc function it is generally preferred to named sec arg style to automatically push and pop the section stack during the scope of the hoc function. ie

h.psection(sec=sec)

The psection section method is different, in that it returns a Python dictionary rather than printing to the screen. It also provides more information, such as reaction-diffusion mechanisms that are present. One could, for example, do

from pprint import pprint
pprint(sec.psection())

The section psection method was added in NEURON 7.6.

With a SectionRef one can, for example,

sr = h.SectionRef(sec=h.dend[2])
sr.root.push(); print(h.secname()); h.pop_section()

or, more compactly and avoiding the modification of the section stack,

sr = h.SectionRef(sec=h.dend[2])
print('%s %s' % (sr.root.name(), h.secname(sec=sr.root)))

Iteration over sections is accomplished with

for s in h.allsec():
  print(s)

sl = h.SectionList(); sl.wholetree()
for s in sl:
  print(s)

In lieu of using a SectionList, one can get the whole tree containing a given section as a Python list via:

tree_secs = my_sec.wholetree()

(The wholetree section method was added in NEURON 7.7.)

Connecting a child section to a parent section uses the connect method using either

childsec.connect(parentsec, parentx, childx)
childsec.connect(parentsegment, childx)

In the first form parentx and childx are optional with default values of 1 and 0 respectively. childx must be 0 or 1 (orientation of the child). Parentx is in the range [0 - 1] but will actually be connected to the center of the parent segment that contains parentx (or exactly at 0 or 1).

sec.cell() returns the cell object that ‘owns’ the section. The return value is None if no object owns the section (a top level section), the instance of the hoc template that created the section, or the python object specified by the named cell parameter when the python section was created.


Segment

Syntax:
seg = section(x)
Description:

A Segment object is obtained from a Section with the function notation where the argument is 0 <= x <= 1 an the segment is the compartment that contains the location x. The x value of the segment is seg.x and the section is seg.sec . From a Segment one can obtain a Mechanism.

To iterate over segments, use for seg in sec: print ("%s(%g)" % (seg.sec.name, seg.x)) This does not include 0 area segments at 0 and 1. For those use for seg in sec.allseg():...


Mechanism

Syntax:
mech = segment.mechname
Description:

A Mechanism object is obtained from a Segment. From a Mechanism one can obtain a range variable. The range variable can also be obtained from the segment by using the hoc range variable name that has the mechanism suffix.

To iterate over density mechanisms, use: for mech in seg: print (mech) To get a python list of point processes in a segment: pplist = seg.point_processes()


HOC accessing Python

Syntax:
nrniv [file.py|file.hoc...]
Description:
The absence of a -python argument causes NEURON to launch with Hoc as the command line interpreter. Python files (or Hoc files) are run with the appropriate interpreter before presenting a Hoc user-interface. From the hoc world any python statement can be executed and anything in the python world can be assigned or evaluated.

nrnpython()
Syntax:
nrnpython("any python statement")
Description:

Executes any python statement. Returns 1 on success; 0 if an exception was raised or if python support is not available.

In particular, python_available = nrnpython("") is 1 (true) if python support is available and 0 (false) if python support is not available.

Example:

nrnpython("import sys")
nrnpython("print(sys.path)")
nrnpython("a = [1,2,3]")
nrnpython("print(a)")
nrnpython("from neuron import h")
nrnpython("h('print PI')")

class PythonObject
Syntax:
p = new PythonObject()
Description:

Accesses any python object. Almost equivalent to HocObject in the python world but because of some hoc syntax limitations, ie. hoc does not allow an object to be a callable function, and top level indices have different semantics, we sometimes need to use a special idiom, ie. the ‘_’ method. Strings and double numbers move back and forth between Python and Hoc (but Python integers, etc. become double values in Hoc, and when they get back to the Python world, they are doubles).

objref p
p = new PythonObject()
nrnpython("ev = lambda arg : eval(arg)") // interprets the string arg as an
                          //expression and returns the value
objref tup
print p.ev("3 + 4")       // prints 7
print p.ev("'hello' + 'world'") // prints helloworld
tup = p.ev("('xyz',2,3)") // tup is a PythonObject wrapping a Python tuple
print tup                 // prints PythonObject[1]
print tup._[2]            // the 2th tuple element is 3
print tup._[0]            // the 0th tuple element is xyz

nrnpython("from neuron import h")   // back in the Python world
nrnpython("print h.tup")  // prints ('xyz', 2, 3)

Note that one needs the ‘_’ method, equivalent to ‘this’, because trying to get at an element through the built-in python method name via

tup.__getitem__(0)

gives the error “TypeError: tuple indices must be integers” since the Hoc 0 argument is a double 0.0 when it gets into Python. It is difficult to pass an integer to a Python function from the hoc world. The only time Hoc doubles appear as integers in Python, is when they are the value of an index. If the index is not an integer, e.g. a string, use the __getitem__ idiom.

objref p
p = new PythonObject()
nrnpython("ev = lambda arg : eval(arg)")
objref d
d = p.ev("{'one':1, 'two':2, 'three':3}")
print d.__getitem__("two")        // prints 2

objref dg
dg = d.__getitem__
print dg._("two")                // prints 2

To assign a value to a python variable that exists in a module use

nrnpython("a = 10")
p = new PythonObject()
p.a = 25
p.a = "hello"
p.a = new Vector(4)
nrnpython("b = []")
p.a = p.b

neuron.hoc.execute()
Syntax:

import neuron

neuron.hoc.execute('any hoc statement')

Description:

Execute any statement or expression using the Hoc interpreter. This is obsolete since the same thing can be accomplished with HocObject with less typing. Note that triple quotes can be used for multiple line statements. A ‘n’ should be escaped as ‘\n’.

hoc.execute('load_file("nrngui.hoc")')

See also

nrnpython()