GUI: missing drive types

[chenghuzi]

In our current GUI code, we have Rhythmic, Poisson and Evoked drives. And if load drives from default.json we got 4 types of them:

However, if we use net = Network(params, add_drives_from_params=True), there are several other types of drives:

And the current GUI, with four drives, generates:

while the latter generates:

[jasmainak]

The extra drives are to keep backwards compatibility with the old GUI. I think you can probably disable them by doing legacy_mode=False. @cjayb @rythorpe any opinions here?

[chenghuzi]

With

import os.path as op

import hnn_core
import matplotlib.pyplot as plt
from hnn_core import (JoblibBackend, MPIBackend, jones_2009_model,
                      simulate_dipole, read_params, Network)
from hnn_core.viz import plot_dipole

hnn_core_root = op.join(op.dirname(hnn_core.__file__))
params_fname = op.join(hnn_core_root, 'param', 'default.json')
params = read_params(params_fname)

# net = jones_2009_model()
net = Network(params, add_drives_from_params=True, legacy_mode=True)

and inserting prints to drives.py

def _add_drives_from_params(net):
    drive_specs = _extract_drive_specs_from_hnn_params(
        net._params, list(net.cell_types.keys()))
    bias_specs = _extract_bias_specs_from_hnn_params(
        net._params, list(net.cell_types.keys()))

    for drive_name in sorted(drive_specs.keys()):  # order matters
        specs = drive_specs[drive_name]
        print(specs['type'], drive_name)

we have as before

However, if I set legacy_mode=False then I get error like this:

[jasmainak]

arghgh ... what if you use jones_2009_model(add_drives_from_params)? This is why zen of Python says:

There should be one-- and preferably only one --obvious way to do it.

I don't think Network(add_drives_from_params) is really meant to be used other than for replicating the old GUI results.

[chenghuzi]

So you mean jones_2009_model is actually a general constructor for all possible network models? If so that will be much easier.

[chenghuzi]

Never mind. You've answered in another thread

[rythorpe]

I'm not sure I understand what's going on here. Why there is a difference between hnn-core and the new GUI when loading default.json, since they both use the same underlying code when handling parameter files?

[jasmainak]

Remember that the GUI always adds some "ghost" drives even if there are no spiking events associated with them. We removed this option in hnn-core over time.

[rythorpe]

Hmm, I still don't understand why the new GUI is doing this. Maybe we can discuss this in our next meeting?

[chenghuzi]

Hmm, I still don't understand why the new GUI is doing this. Maybe we can discuss this in our next meeting?

It's the same now.

Previously it was different as I followed the design that for certain drives we use the `tstop` value for simulation to overwrite the one from the parameter file. So this actually allowed some drives to invoke spikes, though in parameter they're disabled by setting `tstop` to 0.

[chenghuzi]

For all possible drive types, this what I have now from loading default.json:

Please let me know if this is all of them @jasmainak @rythorpe . I checked drives.py and noticed that some drive names actually share the same method under the hood to add drives so I just slightly changed the logic in gui.py to fit this.

[jasmainak]

I think it's all of them!

[rythorpe]

Great, so just to confirm @chenghuzi, loading and simulating default.json using the hnn-core API (i.e., via scripting) vs the GUI now produce exactly the same number/type of drives as well as current dipole output?

[chenghuzi]

Great, so just to confirm @chenghuzi, loading and simulating default.json using the hnn-core API (i.e., via scripting) vs the GUI now produce exactly the same number/type of drives as well as current dipole output?

yes