Refactor

Example/test/configs/l2_3.py

@@ -0,0 +1,54 @@
+from conex.behaviors.neurons.neuron_types.lif_neurons import AELIF, ELIF
+from conex.nn.Config.layer_config import LayerConfig
+from pymonntorch import *
+
+
+class l2_3(LayerConfig):
+    exc_size = (10, 20, 20)
+    exc_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+        "alpha": 0.3,
+        "beta": -0.4,
+        "w_tau": 10,
+        "delta": 0.9,
+        "theta_rh": -42,
+    }
+    exc_neuron_type = AELIF
+    exc_tau_trace = 2.7
+    exc_fire = True
+    exc_dendrite_params = {"distal_provocativeness": 0.5}
+
+    inh_size = (5, 20, 20)
+    inh_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+        "delta": 0.8,
+        "theta_rh": -42,
+    }
+    inh_neuron_type = ELIF
+    inh_tau_trace = 2.7
+    inh_fire = True
+    inh_dendrite_params = {"distal_provocativeness": 0.5}
+
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+
+    exc_inh_weight_init_params = {"mode": "uniform"}
+    exc_inh_structure = "Simple"
+    exc_inh_structure_params = {"current_coef": 3}
+
+    inh_exc_weight_init_params = {"mode": "uniform"}
+    inh_exc_structure = "Simple"
+    inh_exc_structure_params = {"current_coef": 3}
+
+    inh_inh_weight_init_params = {"mode": "uniform"}
+    inh_inh_structure = "Simple"
+    inh_inh_structure_params = {"current_coef": 3}

Example/test/configs/l2_3_l5.py

@@ -0,0 +1,9 @@
+from conex.nn.Config.connection_config import Layer2LayerConnectionConfig
+
+
+class l2_3_l5(Layer2LayerConnectionConfig):
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+    exc_exc_src_pop = "exc_pop"
+    exc_exc_dst_pop = "exc_pop"

Example/test/configs/l4.py

@@ -0,0 +1,49 @@
+from conex.behaviors.neurons.neuron_types.lif_neurons import LIF, ELIF
+from conex.nn.Config.layer_config import LayerConfig
+from pymonntorch import *
+
+
+class l4(LayerConfig):
+    exc_size = (4, 25, 25)
+    exc_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+    }
+    exc_neuron_type = LIF
+    exc_tau_trace = 2.7
+    exc_fire = True
+    exc_dendrite_params = {"distal_provocativeness": 0.5}
+
+    inh_size = (4, 5, 5)
+    inh_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+        "delta": 0.8,
+        "theta_rh": -42,
+    }
+    inh_neuron_type = ELIF
+    inh_tau_trace = 2.7
+    inh_fire = True
+    inh_dendrite_params = {"distal_provocativeness": 0.5}
+
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+
+    exc_inh_weight_init_params = {"mode": "uniform"}
+    exc_inh_structure = "Simple"
+    exc_inh_structure_params = {"current_coef": 3}
+
+    inh_exc_weight_init_params = {"mode": "uniform"}
+    inh_exc_structure = "Simple"
+    inh_exc_structure_params = {"current_coef": 3}
+
+    inh_inh_weight_init_params = {"mode": "uniform"}
+    inh_inh_structure = "Simple"
+    inh_inh_structure_params = {"current_coef": 3}

Example/test/configs/l4_l2_3.py

@@ -0,0 +1,9 @@
+from conex.nn.Config.connection_config import Layer2LayerConnectionConfig
+
+
+class l4_l2_3(Layer2LayerConnectionConfig):
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+    exc_exc_src_pop = "exc_pop"
+    exc_exc_dst_pop = "exc_pop"

Example/test/configs/l5.py

@@ -0,0 +1,51 @@
+from conex.behaviors.neurons.neuron_types.lif_neurons import ELIF
+from conex.nn.Config.layer_config import LayerConfig
+from pymonntorch import *
+
+
+class l5(LayerConfig):
+    exc_size = (3, 5, 5)
+    exc_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+        "delta": 0.8,
+        "theta_rh": -42,
+    }
+    exc_neuron_type = ELIF
+    exc_tau_trace = 2.7
+    exc_fire = True
+    exc_dendrite_params = {"distal_provocativeness": 0.5}
+
+    inh_size = (1, 5, 5)
+    inh_neuron_params = {
+        "R": 20,
+        "tau": 25,
+        "threshold": -37,
+        "v_reset": -75,
+        "v_rest": -65,
+        "delta": 0.8,
+        "theta_rh": -42,
+    }
+    inh_neuron_type = ELIF
+    inh_tau_trace = 2.7
+    inh_fire = True
+    inh_dendrite_params = {"distal_provocativeness": 0.5}
+
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+
+    exc_inh_weight_init_params = {"mode": "uniform"}
+    exc_inh_structure = "Simple"
+    exc_inh_structure_params = {"current_coef": 3}
+
+    inh_exc_weight_init_params = {"mode": "uniform"}
+    inh_exc_structure = "Simple"
+    inh_exc_structure_params = {"current_coef": 3}
+
+    inh_inh_weight_init_params = {"mode": "uniform"}
+    inh_inh_structure = "Simple"
+    inh_inh_structure_params = {"current_coef": 3}

Example/test/configs/l5_l2_3.py

@@ -0,0 +1,9 @@
+from conex.nn.Config.connection_config import Layer2LayerConnectionConfig
+
+
+class l5_l2_3(Layer2LayerConnectionConfig):
+    exc_exc_weight_init_params = {"mode": "uniform"}
+    exc_exc_structure = "Simple"
+    exc_exc_structure_params = {"current_coef": 3}
+    exc_exc_src_pop = "exc_pop"
+    exc_exc_dst_pop = "exc_pop"

Example/test/configs/sens_l4.py

@@ -0,0 +1,9 @@
+from conex.nn.Config.connection_config import Input2LayerConnectionConfig
+
+
+class sens_l4(Input2LayerConnectionConfig):
+    exc_weight_init_params = {"mode": "uniform", "weight_shape": [4, 1, 4, 4]}
+    exc_structure = "Conv2d"
+    exc_structure_params = {"current_coef": 3}
+    exc_dst_pop = "exc_pop"
+    exc_tag = "Proximal"

Example/test/mnist.py

@@ -0,0 +1,104 @@
+from pymonntorch import *
+
+from conex import *
+from conex.nn.timestamps import NEURON_TIMESTAMPS
+
+from conex.helpers.encoders import Poisson
+from conex.helpers.transforms import *
+
+from torchvision import transforms
+from torchvision.datasets import MNIST
+from torch.utils.data import DataLoader
+
+import torch
+
+# parameters
+POISSON_TIME = 30
+POISSON_RATIO = 5 / 30
+MNIST_ROOT = "~/Temp/MNIST/"
+SENSORY_SIZE_HEIGHT = 28
+SENSORY_SIZE_WIDTH = 28  # MNIST's image size
+SENSORY_TRACE_TAU_S = 2.7
+DEVICE = "cuda"
+DTYPE = torch.float16
+
+from configs.l2_3 import *
+from configs.l4 import *
+from configs.l5 import *
+from configs.sens_l4 import *
+from configs.l2_3_l5 import *
+from configs.l4_l2_3 import *
+from configs.l5_l2_3 import *
+
+
+#############################
+# making dataloader
+#############################
+transformation = transforms.Compose(
+    [
+        transforms.ToTensor(),
+        SqueezeTransform(dim=0),
+        Poisson(timesteps=POISSON_TIME, ratio=POISSON_RATIO),
+    ]
+)
+
+dataset = MNIST(root=MNIST_ROOT, train=True, download=True, transform=transformation)
+
+dl = DataLoader(dataset, batch_size=16)
+#############################
+
+
+#############################
+# initializing neocortex
+#############################
+net = Neocortex(settings={"device": DEVICE, "dtype": DTYPE})
+#############################
+
+
+#############################
+# input layer
+#############################
+input_layer = InputLayer(
+    net=net,
+    input_dataloader=dl,
+    sensory_size=NeuronDimension(
+        depth=1, height=SENSORY_SIZE_HEIGHT, width=SENSORY_SIZE_WIDTH
+    ),
+    sensory_trace=SENSORY_TRACE_TAU_S,
+)
+
+#############################
+
+
+#############################
+# making cortical column
+#############################
+cc1 = CorticalColumn(
+    net,
+    L2_3_config=l2_3().make(),
+    L4_config=l4().make(),
+    L5_config=l5().make(),
+    L6_config=None,
+    L6_L4_syn_config=None,
+    L4_L2_3_syn_config=l4_l2_3().make(),
+    L2_3_L5_syn_config=l2_3_l5().make(),
+    L5_L2_3_syn_config=l5_l2_3().make(),
+)
+#############################
+
+
+#############################
+# connect sensory to column
+#############################
+input_layer.connect(
+    cc1,
+    sensory_L4_syn_config=sens_l4().make(),
+    sensory_L6_syn_config=None,
+    location_L6_syn_config=None,
+)
+#############################
+
+
+net.initialize()
+
+net.simulate_iterations(100)

conex/__init__.py

@@ -1,14 +1,17 @@
-from conex.behaviours.network.neuromodulators import *
-from conex.behaviours.network.timestep import *
-from conex.behaviours.network.reward import *
+from conex.behaviors.network.neuromodulators import *
+from conex.behaviors.network.timestep import *
+from conex.behaviors.network.reward import *
 
-from conex.behaviours.neurons.neuron_types.lif_neurons import *
-from conex.behaviours.neurons.sensory.dataset import *
-from conex.behaviours.neurons.specs import *
+from conex.behaviors.neurons.neuron_types.lif_neurons import *
+from conex.behaviors.neurons.setters import *
+from conex.behaviors.neurons.specs import *
 
-from conex.behaviours.synapses.dendrites import *
-from conex.behaviours.synapses.learning import *
-from conex.behaviours.synapses.specs import *
+from conex.behaviors.synapses.dendrites import *
+from conex.behaviors.synapses.learning import *
+from conex.behaviors.synapses.specs import *
+
+from conex.behaviors.layer.dataset import *
 
 from conex.nn.Structure import *
-from conex.nn.Modules import *
+from conex.nn.Modules import *
+from conex.nn.Config import *

conex/behaviors/layer/dataset.py

@@ -0,0 +1,62 @@
+"""
+Behaviors to load datasets 
+"""
+
+from pymonntorch import Behavior
+import torch
+
+
+class SpikeNdDataset(Behavior):
+    """
+    This behavior ease loading dataset as spikes for `InputLayer`.
+    """
+
+    def initialize(self, layer):
+        self.dataloader = self.parameter("dataloader", None, required=True)
+        self.sensory_dimension = self.parameter("N_sensory", 2)
+        self.location_dimension = self.parameter("N_location", 2)
+        self.have_location = self.parameter("have_location", False)
+        self.have_sensory = self.parameter("have_sensory", True)
+        self.have_label = self.parameter("have_label", True)
+        self.loop = self.parameter("loop", True)
+
+        self.data_generator = self._get_data()
+        self.device = layer.device
+
+    def _get_data(self):
+        while self.loop:
+            for batch in self.dataloader:
+                batch_x = batch[0] if self.have_sensory else None
+                batch_loc = batch[self.have_sensory] if self.have_location else None
+                batch_y = batch[-1] if self.have_label else None
+
+                if batch_x is not None:
+                    batch_x = batch_x.to(self.device)
+                    batch_x = batch_x.view(
+                        (-1, *batch_x.shape[-self.sensory_dimension :])
+                    )
+                    num_instance = batch_x.size(0)
+
+                if batch_loc is not None:
+                    batch_loc = batch_loc.to(self.device)
+                    batch_loc = batch_loc.view(
+                        (-1, *batch_loc.shape[-self.location_dimension :])
+                    )
+                    num_instance = batch_loc.size(0)
+                    if batch_x:
+                        assert (
+                            batch_x.size(0) == num_instance
+                        ), "sensory and location should have same number of instances."
+
+                if batch_y is not None:
+                    batch_y = batch_y.to(self.device)
+                    each_instance = num_instance // torch.numel(batch_y)
+
+                for i in range(num_instance):
+                    x = batch_x[i].view((-1,)) if batch_x is not None else None
+                    loc = batch_loc[i].view((-1,)) if batch_loc is not None else None
+                    y = batch_y[i // each_instance] if batch_y is not None else None
+                    yield x, loc, y
+
+    def forward(self, layer):
+        layer.x, layer.loc, layer.y = next(self.data_generator)