Merge pull request #342 from chaoming0625/master

Formalize new styles of neuron and synapse models

README.md

@@ -30,8 +30,8 @@ BrainPy is a flexible, efficient, and extensible framework for computational neu
 
 - **[BrainPy](https://github.com/brainpy/BrainPy)**: The solution for the general-purpose brain dynamics programming. 
 - **[brainpylib](https://github.com/brainpy/brainpylib)**: Efficient operators for the sparse and event-driven computation. 
-- **[BrainPyExamples](https://github.com/brainpy/BrainPyExamples)**: Comprehensive examples of BrainPy computation. 
-- **[brainpy-largescale](https://github.com/NH-NCL/brainpy-largescale)**: One solution for the large-scale brain modeling. 
+- **[brainpy-examples](https://github.com/brainpy/examples)**: Comprehensive examples of BrainPy computation. 
+- **[brainpy-datasets](https://github.com/brainpy/datasets)**: Neuromorphic and Cognitive Datasets for Brain Dynamics Modeling.
 
 
 

brainpy/__init__.py

@@ -58,15 +58,16 @@
                      synapses,  # synaptic dynamics
                      synouts,  # synaptic output
                      synplast,  # synaptic plasticity
-                     syn,
+                     experimental,
                      )
-from brainpy._src.dyn.base import not_pass_sha
+from brainpy._src.dyn.base import not_pass_shared
 from brainpy._src.dyn.base import (DynamicalSystem,
                                    DynamicalSystemNS,
                                    Container as Container,
                                    Sequential as Sequential,
                                    Network as Network,
                                    NeuGroup as NeuGroup,
+                                   NeuGroupNS as NeuGroupNS,
                                    SynConn as SynConn,
                                    SynOut as SynOut,
                                    SynSTP as SynSTP,
@@ -77,8 +78,7 @@
 from brainpy._src.dyn.transform import (NoSharedArg as NoSharedArg,  # transformations
                                         LoopOverTime as LoopOverTime,)
 from brainpy._src.dyn.runners import (DSRunner as DSRunner)  # runner
-from brainpy._src.dyn.context import share
-from brainpy._src.dyn.delay import Delay
+from brainpy._src.dyn.context import share, Delay
 
 
 #  Part 4: Training  #

brainpy/_src/analysis/highdim/slow_points.py

@@ -349,6 +349,7 @@ def f_loss():
     def train(idx):
       gradients, loss = grad_f()
       optimizer.update(gradients if isinstance(gradients, dict) else {'a': gradients})
+      optimizer.lr.step_epoch()
       return loss
 
     def batch_train(start_i, n_batch):

brainpy/_src/dyn/base.py

@@ -50,7 +50,7 @@
 SLICE_VARS = 'slice_vars'
 
 
-def not_pass_sha(func: Callable):
+def not_pass_shared(func: Callable):
   """Label the update function as the one without passing shared arguments.
 
   The original update function explicitly requires shared arguments at the first place::
@@ -610,7 +610,8 @@ def __repr__(self):
     entries = '\n'.join(f'  [{i}] {tools.repr_object(x)}' for i, x in enumerate(self._modules))
     return f'{self.__class__.__name__}(\n{entries}\n)'
 
-  def update(self, s, x) -> ArrayType:
+  @not_pass_shared
+  def update(self, x) -> ArrayType:
     """Update function of a sequential model.
 
     Parameters
@@ -626,12 +627,7 @@ def update(self, s, x) -> ArrayType:
       The output tensor.
     """
     for m in self._modules:
-      if isinstance(m, DynamicalSystemNS):
-        x = m(x)
-      elif isinstance(m, DynamicalSystem):
-        x = m(s, x)
-      else:
-        x = m(x)
+      x = m(x)
     return x
 
 
@@ -665,7 +661,7 @@ def __init__(
                                   mode=mode,
                                   **ds_dict)
 
-  @not_pass_sha
+  @not_pass_shared
   def update(self, *args, **kwargs):
     """Step function of a network.
 
@@ -807,6 +803,11 @@ def __getitem__(self, item):
     return NeuGroupView(target=self, index=item)
 
 
+class NeuGroupNS(NeuGroup):
+  """Base class for neuron group without shared arguments passed."""
+  pass_shared = False
+
+
 class SynConn(DynamicalSystem):
   """Base class to model two-end synaptic connections.