Simulations

From recipe to simulation

To build a simulation the following concepts are needed:

The workflow to build a simulation is to first generate an arbor.domain_decomposition based on the arbor.recipe and arbor.context describing the distribution of the model over the local and distributed hardware resources (see Domain decomposition). Then, the simulation is build using this arbor.domain_decomposition.

import arbor

# Get a communication context (with 4 threads, no GPU)
context = arbor.context(threads=4, gpu_id=None)

# Initialise a recipe of user defined type my_recipe with 100 cells.
n_cells = 100
recipe = my_recipe(n_cells)

# Get a description of the partition of the model over the cores.
decomp = arbor.partition_load_balance(recipe, context)

# Instantiate the simulation.
sim = arbor.simulation(recipe, decomp, context)

# Run the simulation for 2000 ms with time stepping of 0.025 ms
tSim = 2000
dt = 0.025
sim.run(tSim, dt)
class arbor.simulation

The executable form of a model. A simulation is constructed from a recipe, and then used to update and monitor the model state.

Simulations take the following inputs:

The constructor takes

  • an arbor.recipe that describes the model;

  • an arbor.domain_decomposition that describes how the cells in the model are assigned to hardware resources;

  • an arbor.context which is used to execute the simulation.

  • a non-negative int in order to seed the pseudo pandom number generator (optional)

Simulations provide an interface for executing and interacting with the model:

  • Specify what data (spikes, probe results) to record.

  • Advance the model state by running the simulation up to some time point.

  • Retrieve recorded data.

  • Reset simulator state back to initial conditions.

Constructor:

simulation(recipe, domain_decomposition, context, seed)

Initialize the model described by an arbor.recipe, with cells and network distributed according to arbor.domain_decomposition, computational resources described by arbor.context and with a seed value for generating reproducible random numbers (optional, default value: 0).

Updating Model State:

update_connections(recipe)

Rebuild the connection table as described by arbor.recipe::connections_on The recipe must differ only in the return value of its connections_on() when compared to the original recipe used to construct the simulation object.

reset()

Reset the state of the simulation to its initial state. Clears recorded spikes and sample data.

clear_samplers()

Clears recorded spikes and sample data.

run(tfinal, dt)

Run the simulation from current simulation time to tfinal, with maximum time step size dt.

Parameters

  • tfinal – The final simulation time [ms].

  • dt – The time step size [ms].

set_binning_policy(policy, bin_interval)

Set the binning policy for event delivery, and the binning time interval bin_interval if applicable [ms].

Parameters

  • policy – The binning policy of type binning.

  • bin_interval – The binning time interval [ms].

Recording spike data:

record(policy)

Disable or enable recorder of rank-local or global spikes, as determined by the policy.

Parameters

policy – Recording policy of type spike_recording.

spikes()

Return a NumPy structured array of spikes recorded during the course of a simulation. Each spike is represented as a NumPy structured datatype with signature ('source', [('gid', '<u4'), ('index', '<u4')]), ('time', '<f8').

The spikes are sorted in ascending order of spike time, and spikes with the same time are sorted accourding to source gid then index.

Sampling probes:

sample(probeset_id, schedule, policy)

Set up a sampling schedule for the probes associated with the supplied probeset_id of type cell_member. The schedule is any schedule object, as might be used with an event generator — see Recipes for details. The policy is of type sampling_policy. It can be omitted, in which case the sampling will accord with the sampling_policy.lax policy.

The method returns a handle which can be used in turn to retrieve the sampled data from the simulator or to remove the corresponding sampling process.

probe_metadata(probeset_id)

Retrieve probe metadata for the probes associated with the given probeset_id of type cell_member. The result will be a list, with one entry per probe; the specifics of each metadata entry will depend upon the kind of probe in question.

remove_sampler(handle)

Disable the sampling process referenced by the argument handle and remove any associated recorded data.

remove_all_samplers()

Disable all sampling processes and remove any associated recorded data.

samples(handle)

Retrieve a list of sample data associated with the given handle. There will be one entry in the list per probe associated with the probeset id used when the sampling was set up. Each entry is a pair (samples, meta) where meta is the probe metadata as would be returned by probe_metadata(probeset_id), and samples contains the recorded values.

For example, if a probe was placed on a locset describing three positions, samples(handle) will return a list of length 3. In each element, each corresponding to a location in the locset, you’ll find a tuple containing metadata and data, where metadata will be a string describing the location, and data will (usually) be a numpy.ndarray.

An empty list will be returned if no output was recorded for the cell. For simulations that are distributed using MPI, handles associated with non-local cells will return an empty list. It is the responsibility of the caller to gather results over the ranks.

The format of the recorded values (data) will depend upon the specifics of the probe, though generally it will be a NumPy array, with the first column corresponding to sample time and subsequent columns holding the value or values that were sampled from that probe at that time.

progress_banner()

Print a progress bar during simulation, with elapsed milliseconds and percentage of simulation completed.

Types:

class arbor.binning

Enumeration for event time binning policy.

none

No binning policy.

regular

Round time down to multiple of binning interval.

following

Round times down to previous event if within binning interval.

class arbor.spike_recording

Enumeration for spike recording policy.

off

Disable spike recording.

local

Record all generated spikes from cells on this MPI rank.

all

Record all generated spikes from cells on all MPI ranks.

class arbor.sampling_policy

Enumeration for determining sampling policy.

lax

Sampling times may not be exactly as requested in the sampling schedule, but the process of sampling is guaranteed not to disturb the simulation progress or results.

exact

Interrupt the progress of the simulation as required to retrieve probe samples at exactly those times requested by the sampling schedule.

Recording spikes

Spikes are generated by various sources, including detectors and spike source cells, but by default, spikes are not recorded. Recording is enabled with the simulation.record() method, which takes a single argument instructing the simulation object to record no spikes, all locally generated spikes, or all spikes generated by any MPI rank.

Spikes recorded during a simulation are returned as a NumPy structured datatype with two fields, source and time. The source field itself is a structured datatype with two fields, gid and index, identifying the threshold detector that generated the spike.

Note

The spikes returned by simulation.record() are sorted in ascending order of spike time. Spikes that have the same spike time are sorted in ascending order of gid and local index of the spike source.

import arbor

# Instantiate the simulation.
sim = arbor.simulation(recipe, decomp, context)

# Direct the simulation to record all spikes, which will record all spikes
# across multiple MPI ranks in distrubuted simulation.
# To only record spikes from the local MPI rank, use arbor.spike_recording.local
sim.record(arbor.spike_recording.all)

# Run the simulation for 2000 ms with time stepping of 0.025 ms
tSim = 2000
dt = 0.025
sim.run(tSim, dt)

# Print the spikes and according spike time
for s in sim.spikes():
    print(s)

>>> ((0,0), 2.15168)
>>> ((1,0), 14.5235)
>>> ((2,0), 26.9051)
>>> ((3,0), 39.4083)
>>> ((4,0), 51.9081)
>>> ((5,0), 64.2902)
>>> ((6,0), 76.7706)
>>> ((7,0), 89.1529)
>>> ((8,0), 101.641)
>>> ((9,0), 114.125)

Recording samples

Procedure

There are three parts to the process of recording cell data over a simulation.

  1. Describing what to measure.

    The recipe object must provide a method recipe.probes() that returns a list of probeset addresses for the cell with a given gid. The kth element of the list corresponds to the probeset id (gid, k).

    Each probeset address is an opaque object describing what to measure and where, and each cell kind will have its own set of functions for generating valid address specifications. Possible cable cell probes are described in the cable cell documentation: Cable cell probing and sampling.

  2. Instructing the simulator to record data.

    Recording is set up with the method simulation.sample() as described above. It returns a handle that is used to retrieve the recorded data after simulation.

  3. Retrieve recorded data.

    The method simulation.samples() takes a handle and returns the recorded data as a list, with one entry for each probe associated with the probeset id that was used in step 2 above. Each entry will be a tuple (data, meta) where meta is the metadata associated with the probe, and data contains all the data sampled on that probe over the course of the simulation.

    The contents of data will depend upon the specifics of the probe, but note:

    1. The object type and structure of data is fully determined by the metadata.

    2. All currently implemented probes return data that is a NumPy array, with one row per sample, first column being sample time, and the remaining columns containing the corresponding data.

Example

import arbor

# [... define recipe, decomposition, context ... ]
# Initialize simulation:

sim = arbor.simulation(recipe, decomp, context)

# Sample probeset id (0, 0) (first probeset id on cell 0) every 0.1 ms with exact sample timing:

handle = sim.sample((0, 0), arbor.regular_schedule(0.1), arbor.sampling_policy.exact)

# Run simulation and retrieve sample data from the first probe associated with the handle.

sim.run(tfinal=3, dt=0.1)
data, meta = sim.samples(handle)[0]
print(data)

>>> [[  0.         -50.        ]
>>>  [  0.1        -55.14412111]
>>>  [  0.2        -59.17057625]
>>>  [  0.3        -62.58417912]
>>>  [  0.4        -65.47040168]
>>>  [  0.5        -67.80222861]
>>>  [  0.6        -15.18191623]
>>>  [  0.7         27.21110919]
>>>  [  0.8         48.74665099]
>>>  [  0.9         48.3515727 ]
>>>  [  1.          41.08435987]
>>>  [  1.1         33.53571111]
>>>  [  1.2         26.55165892]
>>>  [  1.3         20.16421752]
>>>  [  1.4         14.37227532]
>>>  [  1.5          9.16209063]
>>>  [  1.6          4.50159342]
>>>  [  1.7          0.34809083]
>>>  [  1.8         -3.3436289 ]
>>>  [  1.9         -6.61665687]
>>>  [  2.          -9.51020525]
>>>  [  2.1        -12.05947812]
>>>  [  2.2        -14.29623969]
>>>  [  2.3        -16.24953688]
>>>  [  2.4        -17.94631322]
>>>  [  2.5        -19.41182385]
>>>  [  2.6        -52.19519009]
>>>  [  2.7        -62.53349949]
>>>  [  2.8        -69.22068995]
>>>  [  2.9        -73.41691825]]