Cannot simulate PointNet network with custom set_syn_weight function

Hello!

I built a pointnet network for which I used add_properties to define the delays with a given rule. By doing that, i deleted the "nsyn" section in the edge h5 file (which is compensated by adding multiple time a connection in the edge h5 file according to its number of synapses). However, during the simulation of the config file I get this error:

Traceback (most recent call last):
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/BMTK_Simulations/grid_searching_l4l4.py", line 31, in <module>
    sim = pointnet.PointSimulator.from_config(configure, graph)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointsimulator.py", line 243, in from_config
    graph.build_recurrent_edges()
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointnetwork.py", line 167, in build_recurrent_edges
    for edge in edge_pop.get_edges():
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/core/sonata_reader/network_reader.py", line 191, in get_edges
    for edge in edge_adaptor.get_batches(edge_group):
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/sonata_adaptors.py", line 335, in get_batches
    type_params[edge_id]['weight'] = weight_fnc(edges, source_nodes, target_nodes)
                                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/default_setters/synaptic_weights.py", line 7, in default_weight_fnc
    return np.array(edge_props['syn_weight']*edge_props.nsyns)
                                             ^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/pandas/core/generic.py", line 6296, in __getattr__
    return object.__getattribute__(self, name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'DataFrame' object has no attribute 'nsyns'
srun: error: node105: task 0: Exited with exit code 1

I tried to overwrite the default set_syn_weight function by building my own set_syn_weight function (which I will also need to use in the future since I want to do a grid searching in order to find an optimal synaptic weight according to a targeted output frequency):

list_grid=[1,2,3,4]
grid_search = {
    'exc':list_grid,
    'in':[0]*len(list_grid)
}

@synaptic_weight
def set_syn_weight(edge_props, source_nodes, target_nodes):

    target_type = target_nodes['cell_line'].unique()[0]
    new_syn_weight = grid_search[target_type][current_itr]
    
    weights_table[target_type].append(new_syn_weight)
    
    return np.array(new_syn_weight*edge_props.nsyns)

But I get this error:

Traceback (most recent call last):
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/BMTK_Simulations/grid_searching_l4l4.py", line 95, in <module>
    sim = pointnet.PointSimulator.from_config(configure, graph)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointsimulator.py", line 243, in from_config
    graph.build_recurrent_edges()
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointnetwork.py", line 173, in build_recurrent_edges
    self._nest_connect(nest_srcs, nest_trgs, conn_spec='one_to_one', syn_spec=edge.nest_params)
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointnetwork.py", line 232, in _nest_connect
    nest.Connect(nest_srcs, nest_trgs, conn_spec=conn_spec, syn_spec=syn_spec)
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/nest/ll_api.py", line 229, in stack_checker_func
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/nest/lib/hl_api_connections.py", line 257, in Connect
    connect_arrays(pre, post, weights, delays, synapse_model, syn_param_keys, syn_param_values)
  File "pynestkernel.pyx", line 300, in pynestkernel.NESTEngine.connect_arrays
TypeError: weights must be a 1-dimensional NumPy array
srun: error: node200: task 0: Exited with exit code 1

I changed the function with:

return np.array(new_syn_weight*pd.Series([1]))

But then again I get this error:

Traceback (most recent call last):
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/BMTK_Simulations/grid_searching_l4l4.py", line 95, in <module>
    sim = pointnet.PointSimulator.from_config(configure, graph)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointsimulator.py", line 243, in from_config
    graph.build_recurrent_edges()
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointnetwork.py", line 173, in build_recurrent_edges
    self._nest_connect(nest_srcs, nest_trgs, conn_spec='one_to_one', syn_spec=edge.nest_params)
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/bmtk/simulator/pointnet/pointnetwork.py", line 232, in _nest_connect
    nest.Connect(nest_srcs, nest_trgs, conn_spec=conn_spec, syn_spec=syn_spec)
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/nest/ll_api.py", line 229, in stack_checker_func
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/nest/lib/hl_api_connections.py", line 214, in Connect
    processed_syn_spec = _process_syn_spec(
                         ^^^^^^^^^^^^^^^^^^
  File "/gpfs/users/calicem/.conda/envs/ENV_NEST33/lib/python3.11/site-packages/nest/lib/hl_api_connection_helpers.py", line 89, in _process_syn_spec
    raise kernel.NESTError(
nest.lib.hl_api_exceptions.NESTErrors.NESTError: 'weight' has to be an array of dimension 299430.
srun: error: node200: task 0: Exited with exit code 1

How can I build my own set_syn_weight function and avoid this issue ?

I am using NEST 3.3 and BMTK 1.0.8.

Thanks in advance for your help,
Margaux

Start with pointnet/default_setters/synaptic_weights.py and pointnet/sonata_adaptors.py, then trace how the custom function's edge_props result reaches pointnet/pointnetwork.py and NEST Connect. Check the expected edge properties and weight array shape against the reported 299430 connections; done means a custom set_syn_weight function runs successfully with the provided edge data.