Support for activation vector generation given source and target node

actflowcomp/__init__.py

@@ -1,4 +1,4 @@
 from .actflowcalc import *
 from .actflowtest import *
 from .noiseceilingcalc import *
-
+from .actflow_sourcetarget import *

actflowcomp/actflow_sourcetarget.py

@@ -0,0 +1,57 @@
+import numpy as np
+
+def actflow_sourcetarget(actvect_sources, nodelist_sources, nodelist_targets, fc_mat, separate_activations_bytarget=False, transfer_func=None):
+    """
+
+    Function to run activity flow mapping algorithm
+
+    actvect_sources: node vector with activation values for source regions
+    nodelist_sources: list of source regions
+    nodelist_targets: list of target regions
+    fcMat: target nodes x source node matrix with connectivity values
+    separate_activations_bytarget: indicates if the input actVect matrix has a separate activation vector for each target (to-be-predicted) node (e.g., for the locally-non-circular approach)
+    transfer_func: The transfer function to apply to the outputs of all source regions. Assumes observed time series are primarily driven by inputs (e.g., local field potentials), such that the source time series need to be converted from inputs to outputs via a transfer function. Default is 'None', which specifies a linear transfer function wherein the output is the same as the input.
+    """
+
+    numSourceNodes = len(nodelist_sources)
+    numTargetNodes = len(nodelist_targets)
+    actPredVector  = np.zeros((numTargetNodes,))
+    sourceRegions  = list(range(numSourceNodes))
+
+    if transfer_func is None:
+
+        if separate_activations_bytarget:
+            for targetRegion in range(numTargetNodes):
+                actPredVector[targetRegion]=np.sum(actvect_sources[targetRegion,sourceRegions]*fc_mat[targetRegion,sourceRegions])
+        else:
+            for targetRegion in range(numTargetNodes):
+                actPredVector[targetRegion]=np.sum(actvect_sources[sourceRegions]*fc_mat[targetRegion,sourceRegions])
+        return actPredVector
+
+    else:
+
+        if separate_activations_bytarget:
+
+            for targetRegion in range(numTargetNodes):
+
+                inputActVect=transfer_function(actvect_sources[targetRegion,sourceRegions],transfer_func=transfer_func)
+                actPredVector[targetRegion]=np.sum(inputActVect*fc_mat[targetRegion,sourceRegions])
+        else:
+            for targetRegion in range(numTargetNodes):
+
+                inputActVect=transfer_function(actvect_sources[sourceRegions],transfer_func=transfer_func)
+                actPredVector[targetRegion]=np.sum(inputActVect*fc_mat[targetRegion,sourceRegions])
+        return actPredVector
+
+
+#Define input transfer function
+def transfer_function(activity, transfer_func='linear', threshold=0, a=1):
+    if transfer_func == 'linear':
+        return activity
+    elif transfer_func == 'relu':
+        return activity*(activity>threshold)
+    elif transfer_func == 'sigmoid':
+        return 1 / (1 + np.exp(-activity))
+    elif transfer_func == 'logit':
+        return (1/a)*np.log(activity/(1-activity))
+