-
Notifications
You must be signed in to change notification settings - Fork 86
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
added rl multinode synchronization environment
- Loading branch information
Showing
3 changed files
with
520 additions
and
0 deletions.
There are no files selected for viewing
Large diffs are not rendered by default.
Oops, something went wrong.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
143 changes: 143 additions & 0 deletions
143
neurolib/control/reinforcement_learning/environments/synchronization.py
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,143 @@ | ||
import numpy as np | ||
|
||
import gymnasium as gym | ||
from gymnasium import spaces | ||
|
||
from neurolib.models.wc import WCModel | ||
|
||
|
||
class SynchronizationEnv(gym.Env): | ||
|
||
def __init__( | ||
self, | ||
duration=200, | ||
dt=0.1, | ||
observation_window=300, # number of observed integration steps | ||
target="sync", | ||
l1_control_strength_loss_scale=1.0, | ||
l2_control_strength_loss_scale=1.0, | ||
): | ||
self.duration = duration | ||
self.dt = dt | ||
self.observation_window = observation_window | ||
self.target = target | ||
self.l1_control_strength_loss_scale = l1_control_strength_loss_scale | ||
self.l2_control_strength_loss_scale = l2_control_strength_loss_scale | ||
|
||
assert target in ("sync", "desync") | ||
if target == "sync": | ||
self.exc_ext_baseline = 1.6 # starts in desync | ||
elif target == "desync": | ||
self.exc_ext_baseline = 1.0 # starts in sync | ||
self.inh_ext_baseline = 0.4 | ||
self.coupling = 0.8 | ||
self.N = 6 | ||
self.cmat = np.array( | ||
[ | ||
[0.0, 1.0, 0.0, 0.0, 0.0, 1.0], | ||
[1.0, 0.0, 1.0, 0.0, 1.0, 0.0], | ||
[0.0, 0.0, 0.0, 0.0, 0.0, 1.0], | ||
[0.0, 0.0, 1.0, 0.0, 0.0, 1.0], | ||
[1.0, 1.0, 0.0, 1.0, 0.0, 1.0], | ||
[0.0, 1.0, 1.0, 0.0, 0.0, 0.0], | ||
] | ||
) | ||
self.dmat = np.array( | ||
[ | ||
[0.0, 12.0, 0.0, 0.0, 0.0, 8.0], | ||
[8.0, 0.0, 13.0, 0.0, 1.0, 0.0], | ||
[0.0, 0.0, 0.0, 0.0, 0.0, 9.0], | ||
[0.0, 0.0, 4.0, 0.0, 0.0, 11.0], | ||
[5.0, 17.0, 0.0, 14.0, 0.0, 18.0], | ||
[0.0, 0.0, 3.0, 0.0, 0.0, 0.0], | ||
] | ||
) | ||
assert np.max(self.dmat / dt) < self.observation_window | ||
|
||
# numerically determined dominant freq in sync state, corresponds to ~19ms period | ||
self.oscillation_freq = 0.052 | ||
""" | ||
def get_oscillation_freq(data, dt): | ||
ps = np.abs(np.fft.fft(data, axis=-1)) | ||
ps[:, 0] = 0. | ||
freqs = np.fft.fftfreq(round(ps.shape[1]*dt)) | ||
return freqs[scipy.stats.mode(ps.argmax(axis=-1)).mode] | ||
""" | ||
|
||
self.model = WCModel(Cmat=self.cmat, Dmat=self.dmat) | ||
self.model.params["dt"] = self.dt | ||
self.model.params["K_gl"] = self.coupling | ||
self.model.params["exc_ext_baseline"] = self.exc_ext_baseline | ||
self.model.params["inh_ext_baseline"] = self.inh_ext_baseline | ||
self.model.params["duration"] = self.dt # one step at a time | ||
|
||
self.n_steps = round(self.duration / self.dt) | ||
|
||
# TODO | ||
self.observation_space = spaces.Dict( | ||
{ | ||
"exc": spaces.Box(0, 1, shape=(self.N, self.observation_window), dtype=float), | ||
"inh": spaces.Box(0, 1, shape=(self.N, self.observation_window), dtype=float), | ||
} | ||
) | ||
|
||
self.action_space = spaces.Tuple( | ||
( | ||
spaces.Box(-5, 5, shape=(self.N,), dtype=float), # exc | ||
spaces.Box(-5, 5, shape=(self.N,), dtype=float), # inh | ||
) | ||
) | ||
|
||
def _get_obs(self): | ||
return { | ||
"exc": self.model.exc[:, -self.observation_window :], | ||
"inh": self.model.inh[:, -self.observation_window :], | ||
} | ||
|
||
def _get_info(self): | ||
return {"t": self.t_i * self.dt} | ||
|
||
def reset(self, seed=None, options=None): | ||
super().reset(seed=seed, options=options) | ||
self.t_i = 0 | ||
self.model.clearModelState() | ||
|
||
self.model.params["duration"] = self.observation_window * self.dt | ||
self.model.run(continue_run=True, append_outputs=True) | ||
self.model.params["duration"] = self.dt # one step at a time | ||
|
||
observation = self._get_obs() | ||
info = self._get_info() | ||
return observation, info | ||
|
||
def synchronization(self, data): | ||
summed = data.sum(0) | ||
ps = np.abs(np.fft.fft(summed)) | ||
freqs = np.fft.fftfreq(round(summed.shape[0] * self.dt)) | ||
return ps[np.argmin(np.abs(freqs - self.oscillation_freq))] | ||
|
||
def _reward(self, obs, action): | ||
if self.target == "sync": | ||
control_reward = self.synchronization(obs["exc"]) | ||
elif self.target == "desync": | ||
control_reward = -1 * self.synchronization(obs["exc"]) | ||
control_strength_loss = np.abs(action).sum() * self.l1_control_strength_loss_scale | ||
control_strength_loss += np.sqrt(np.sum(np.square(action))) * self.l2_control_strength_loss_scale | ||
return control_reward - control_strength_loss | ||
|
||
def step(self, action): | ||
assert self.action_space.contains(action) | ||
exc, inh = action | ||
self.model.params["exc_ext"] = np.array(exc) | ||
self.model.params["inh_ext"] = np.array(inh) | ||
self.model.run(continue_run=True, append_outputs=True) | ||
|
||
observation = self._get_obs() | ||
|
||
reward = self._reward(observation, action) | ||
|
||
self.t_i += 1 | ||
terminated = self.t_i >= self.n_steps | ||
info = self._get_info() | ||
|
||
return observation, reward, terminated, False, info |