margulies · November 29, 2016 21:07
diff --git a/macaque_embedding.py b/macaque_embedding.py
 import numpy as np
 import nibabel as nib # git clone --branch enh/cifti2 https://github.com/satra/nibabel.git
 from sklearn.metrics import pairwise_distances

 # install mapalign: (comment if already installed)
 ! pip install git+https://github.com/satra/mapalign
 from mapalign import embed

 # Load data and Fisher's z-to-r transform
 # load connectivity matrix as numpy array
 data = #
 # If input matrix is already r-to-z transfromed, include next line.
 # If input matrix is r-values, comment out.
 dcon = np.tanh(data)

 # Get number of nodes
 N = dcon.shape[0]

 # Generate percentile thresholds for 90th percentile
 perc = np.array([np.percentile(x, 90) for x in dcon])

 # Threshold each row of the matrix by setting values below 90th percentile to 0
 for i in range(dcon.shape[0]):
  print "Row %d" % i
  dcon[i, dcon[i,:] < perc[i]] = 0

 # Check for minimum value
 print "Minimum value is %f" % dcon.min()

 # The negative values are very small, but we need to know how many nodes have negative values
 # Count negative values per row
 neg_values = np.array([sum(dcon[i,:] < 0) for i in range(N)])
 print "Negative values occur in %d rows" % sum(neg_values > 0)

 # Since there are only 23 vertices with total of 5000 very small negative values, we set these to zero
 dcon[dcon < 0] = 0

 # Now we are dealing with sparse vectors. Cosine similarity is used as affinity metric
 aff = 1 - pairwise_distances(dcon, metric = 'cosine')

 # Save affinity matrix (optional)
 # uncomment to save:
 # np.save('gradient_data/conn_matrices/cosine_affinity.npy', aff)

 # Load affinitity matrix
 # if saved, uncomment:
 # aff = np.load('gradient_data/conn_matrices/cosine_affinity.npy')

 emb, res = embed.compute_diffusion_map(aff, alpha = 0.5)

 # Save results
 np.save('mbedding_dense_emb.npy', emb)
 np.save('embedding_dense_res.npy', res)

 # variable 'emb' connectains the connectivity gradients in order of variance explained
	import numpy as np
	import nibabel as nib # git clone --branch enh/cifti2 https://github.com/satra/nibabel.git
	from sklearn.metrics import pairwise_distances

	# install mapalign: (comment if already installed)
	! pip install git+https://github.com/satra/mapalign
	from mapalign import embed

	# Load data and Fisher's z-to-r transform
	# load connectivity matrix as numpy array
	data = #
	# If input matrix is already r-to-z transfromed, include next line.
	# If input matrix is r-values, comment out.
	dcon = np.tanh(data)

	# Get number of nodes
	N = dcon.shape[0]

	# Generate percentile thresholds for 90th percentile
	perc = np.array([np.percentile(x, 90) for x in dcon])

	# Threshold each row of the matrix by setting values below 90th percentile to 0
	for i in range(dcon.shape[0]):
	print "Row %d" % i
	dcon[i, dcon[i,:] < perc[i]] = 0

	# Check for minimum value
	print "Minimum value is %f" % dcon.min()

	# The negative values are very small, but we need to know how many nodes have negative values
	# Count negative values per row
	neg_values = np.array([sum(dcon[i,:] < 0) for i in range(N)])
	print "Negative values occur in %d rows" % sum(neg_values > 0)

	# Since there are only 23 vertices with total of 5000 very small negative values, we set these to zero
	dcon[dcon < 0] = 0

	# Now we are dealing with sparse vectors. Cosine similarity is used as affinity metric
	aff = 1 - pairwise_distances(dcon, metric = 'cosine')

	# Save affinity matrix (optional)
	# uncomment to save:
	# np.save('gradient_data/conn_matrices/cosine_affinity.npy', aff)

	# Load affinitity matrix
	# if saved, uncomment:
	# aff = np.load('gradient_data/conn_matrices/cosine_affinity.npy')

	emb, res = embed.compute_diffusion_map(aff, alpha = 0.5)

	# Save results
	np.save('mbedding_dense_emb.npy', emb)
	np.save('embedding_dense_res.npy', res)

	# variable 'emb' connectains the connectivity gradients in order of variance explained