Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[REFACTOR] Black and Isort #24

Merged
merged 1 commit into from
Mar 26, 2023
Merged

[REFACTOR] Black and Isort #24

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
104 changes: 51 additions & 53 deletions clustintime/Clustering.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,25 +4,26 @@
Clustering library for clustintime
"""
import os
import random

import infomap
import networkx as nx # creation, manipulation and study of the structure, dynamics and functions of complex networks
import nibabel as nib
import numpy as np
import pandas as pd
import random

import clustintime.Visualization as vis
import clustintime.Processing as proc
from sklearn.cluster import KMeans
from sklearn.cluster import DBSCAN
from sklearn.cluster import OPTICS
from sklearn.cluster import AffinityPropagation
from sklearn.cluster import MeanShift
from sklearn.cluster import AgglomerativeClustering
from networkx.algorithms import community
from community import community_louvain
from networkx.algorithms import community
from sklearn.cluster import (
DBSCAN,
OPTICS,
AffinityPropagation,
AgglomerativeClustering,
KMeans,
MeanShift,
)

import clustintime.Processing as proc
import clustintime.Visualization as vis


def generate_maps(labels, directory, data, masker, prefix):
Expand Down Expand Up @@ -56,14 +57,15 @@ def generate_maps(labels, directory, data, masker, prefix):

for map_idx in range(len(unique)):
mean_img = np.mean(data[labels == map_idx + 1], axis=0)
if mean_img.min()/mean_img.max() < 0.9:
mean_img = mean_img/mean_img.max()
if mean_img.min() / mean_img.max() < 0.9:
mean_img = mean_img / mean_img.max()
mean_img_3d = masker.inverse_transform(
mean_img
) # Transform the averaged image into a 3D image

nib.save(
mean_img_3d, os.path.join(directory, f"{directory}/{prefix}_cluster_{map_idx+1}.nii.gz")
mean_img_3d,
os.path.join(directory, f"{directory}/{prefix}_cluster_{map_idx+1}.nii.gz"),
)


Expand Down Expand Up @@ -94,12 +96,13 @@ def findCommunities(G):
nx.set_node_attributes(G, values=communities, name="community")
return communities


def consensus(corr_map, indexes, nscans, n_clusters, algorithm, thr):
"""
This algorithm samples the data and clusters it with a defined algorithm. With the results of each cluster, it creates a consensus matrix.
This algorithm samples the data and clusters it with a defined algorithm. With the results of each cluster, it creates a consensus matrix.
The consensus matrix is then clustered a hundred times. If the results are the same in every run, that will be the returned labels.
If the results are not unanimous, the algorithm will return to the sampling step.
If the results are not unanimous, the algorithm will return to the sampling step.

Parameters
----------
corr_map : matrix
Expand All @@ -121,38 +124,38 @@ def consensus(corr_map, indexes, nscans, n_clusters, algorithm, thr):
I_sum = np.zeros([npoints, npoints])
while 1:
for i in range(100):
sampling = np.sort(random.sample(range(npoints), round(npoints*0.6)))
I = pd.DataFrame([0]*npoints)
sampling = np.sort(random.sample(range(npoints), round(npoints * 0.6)))
I = pd.DataFrame([0] * npoints)
I[0][sampling] = 1
I_sum = I_sum + np.dot(I,np.transpose(I))
data_sampled = corr_map[sampling,:][:,sampling]
I_sum = I_sum + np.dot(I, np.transpose(I))
data_sampled = corr_map[sampling, :][:, sampling]
if algorithm == Info_Map or algorithm == Greedy_Mod or algorithm == Louvain:
corr_mat, idx = algorithm(data_sampled, indexes, thr, nscans)
corr_mat, idx = algorithm(data_sampled, indexes, thr, nscans)
idx = np.transpose(
pd.DataFrame([idx, sampling])
) # Create a vector that combines the previous indexes and the labels
else:
idx = algorithm(data_sampled, indexes, nscans, n_clusters)
) # Create a vector that combines the previous indexes and the labels
else:
idx = algorithm(data_sampled, indexes, nscans, n_clusters)
idx = np.transpose(
pd.DataFrame([idx, sampling])
) # Create a vector that combines the previous indexes and the labels
) # Create a vector that combines the previous indexes and the labels
idx = idx.set_index(1)
idx = idx[np.logical_not(np.isnan(idx[0]))]
labels = np.array([0]*npoints)
labels = np.array([0] * npoints)
labels[sampling] = idx[0]
M = proc.compute_connectivity_matrix(npoints, labels)
M = proc.compute_connectivity_matrix(npoints, labels)
M_sum = M_sum + M
Consensus = np.divide(M_sum,I_sum)
Consensus = np.divide(M_sum, I_sum)
Consensus[Consensus < proc.find_threshold_bfs(Consensus)] = 0
final_labels = algorithm(Consensus, indexes, 0, nscans)
thr = proc.find_threshold_bfs(Consensus)
Consensus[Consensus <= thr] = 0
aux = proc.compute_connectivity_matrix(npoints, final_labels[1])
aux = proc.compute_connectivity_matrix(npoints, final_labels[1])
boolean = True
for i in range(100):
labels = algorithm(corr_map = Consensus,indexes=indexes,thr=thr, nscans=npoints)
labels = algorithm(corr_map=Consensus, indexes=indexes, thr=thr, nscans=npoints)
connect = proc.compute_connectivity_matrix(npoints, labels[1])

if np.array_equal(aux, connect) == False:
boolean = False
break
Expand All @@ -161,7 +164,7 @@ def consensus(corr_map, indexes, nscans, n_clusters, algorithm, thr):
return labels[1]


def K_Means(corr_map, indexes, nscans, n_clusters, seed = 0):
def K_Means(corr_map, indexes, nscans, n_clusters, seed=0):
"""
K-Means uses a pre-stablished number of centroids and iterations defined by the user.
The algorithms places the centroids at random locations (real or imaginary, that represent the centre of the cluster) and then allocates each data point to the nearest cluster.
Expand Down Expand Up @@ -192,7 +195,7 @@ def K_Means(corr_map, indexes, nscans, n_clusters, seed = 0):

print(" ")

KM = KMeans(n_clusters=n_clusters, random_state = seed)
KM = KMeans(n_clusters=n_clusters, random_state=seed)
labels = KM.fit_predict(corr_map)

labels = np.transpose(
Expand All @@ -205,14 +208,16 @@ def K_Means(corr_map, indexes, nscans, n_clusters, seed = 0):
# assign to each timepoint their label

for i in labels.index:
i=int(i)
i = int(i)
final_labels[i] = labels[0][i] + 1
print("K-Means applied!")

return final_labels


def Agglomerative_Clustering(corr_map, indexes, nscans, n_clusters, affinity = 'euclidean', linkage = 'ward'):
def Agglomerative_Clustering(
corr_map, indexes, nscans, n_clusters, affinity="euclidean", linkage="ward"
):
"""
Agglomerative Clustering recursively merges the pair of clusters that minimally increases a given linkage distance.

Expand All @@ -227,8 +232,8 @@ def Agglomerative_Clustering(corr_map, indexes, nscans, n_clusters, affinity = '
n_clusters : int
Number of clusters.
affinity : str, optional
Metric used to compute the linkage. Can be `euclidean`, `l1`, `l2`, `manhattan`, `cosine`, or `precomputed`.
If linkage is `ward`, only `euclidean` is accepted.
Metric used to compute the linkage. Can be `euclidean`, `l1`, `l2`, `manhattan`, `cosine`, or `precomputed`.
If linkage is `ward`, only `euclidean` is accepted.
If `precomputed`, a distance matrix (instead of a similarity matrix) is needed as input for the fit method.
The default is `euclidean`
linkage : str, optional:
Expand All @@ -247,7 +252,7 @@ def Agglomerative_Clustering(corr_map, indexes, nscans, n_clusters, affinity = '

print(" ")

AG = AgglomerativeClustering(n_clusters = n_clusters, affinity = affinity, linkage = linkage)
AG = AgglomerativeClustering(n_clusters=n_clusters, affinity=affinity, linkage=linkage)
labels = AG.fit_predict(corr_map)

labels = np.transpose(
Expand All @@ -260,14 +265,13 @@ def Agglomerative_Clustering(corr_map, indexes, nscans, n_clusters, affinity = '
# assign to each timepoint their label

for i in labels.index:
i=int(i)
i = int(i)
final_labels[i] = labels[0][i] + 1
print("Agglomerative Clustering applied!")

return final_labels



def Info_Map(corr_map, indexes, thr, nscans):
"""
InfoMap uses information theory to find communities. In particular, it employs the Huffman code to understand the flow of information within a graph. This code assigns a prefix to each node, then a prefix to each community.
Expand Down Expand Up @@ -298,7 +302,6 @@ def Info_Map(corr_map, indexes, thr, nscans):
print("Applying InfoMap... ")
print(" ")


corr_map = proc.thr_index(corr_map, thr)
corr_smooth_binary = corr_map != 0 # Find all the voxels with correlation

Expand All @@ -322,6 +325,7 @@ def Info_Map(corr_map, indexes, thr, nscans):

return corr_smooth_binary, final_labels


def Louvain(corr_map, indexes, thr, nscans):
"""
Louvain's algorithm maximises modularity and implements an extra step to ensure community properties in the network.
Expand All @@ -336,7 +340,7 @@ def Louvain(corr_map, indexes, thr, nscans):
Percentile threshold for the binarization.
nscans : int
Number of scans.

Returns
-------
corr_map : matrix
Expand All @@ -349,12 +353,10 @@ def Louvain(corr_map, indexes, thr, nscans):
print("Applying Louvain... ")
print(" ")



# compute the best partition

corr_map = proc.thr_index(corr_map, thr)

corr_smooth_binary = corr_map != 0 # Find all the voxels with correlation

G = nx.from_numpy_matrix(corr_smooth_binary) # Again the binary
Expand All @@ -375,9 +377,9 @@ def Louvain(corr_map, indexes, thr, nscans):

print("Louvain applied")


return corr_smooth_binary, final_labels


def Greedy_Mod(corr_map, indexes, thr, nscans):
"""
Greedy modularity maximises modularity.
Expand Down Expand Up @@ -405,12 +407,10 @@ def Greedy_Mod(corr_map, indexes, thr, nscans):
print("Applying Greedy Modularity... ")
print(" ")



# compute the best partition

corr_map = proc.thr_index(corr_map, thr)

corr_smooth_binary = corr_map != 0 # Find all the voxels with correlation

G = nx.from_numpy_matrix(corr_smooth_binary) # Again the binary
Expand All @@ -431,6 +431,4 @@ def Greedy_Mod(corr_map, indexes, thr, nscans):

print("Greedy Modularity applied")


return corr_smooth_binary, final_labels

Loading