substituting sklearn.AgglomerativeClustering for fastclustering and s…

…cipy's dendrogram for ROW ONLY

src/HierarchicalClustering.py

@@ -8,14 +8,16 @@
 import os
 from hc_functions import *
 import seaborn as sns
-from sklearn.cluster import AgglomerativeClustering
+from sklearn.cluster import AgglomerativeClustering  # We are not using this anymore
 import matplotlib as mpl
 import humanfriendly
 import datetime
 from inspect import currentframe
 tasklib_path = os.path.dirname(os.path.realpath(sys.argv[0]))
-# mpl.use('Agg')"
+# mpl.use('Agg')
 sns.set_style("white")
+import fastcluster
+from scipy.spatial.distance import pdist
 
 DEBUG = True
 
@@ -79,14 +81,26 @@ def log(text, line_number='?', debug=True):
 if row_distance_metric != 'No_row_clustering':
     gtr_companion = True
 
-    # Set Sklearn's clustering model parameters
-    row_model = AgglomerativeClustering(linkage=linkage_dic[clustering_method], n_clusters=2,
-                                        affinity=str2func[row_distance_metric])
-    # fit Sklearn's clustering model
-    row_model.fit(data)
-    row_tree = make_tree(row_model)
-    order_of_rows = order_leaves(row_model, tree=row_tree, data=data,
-                                 dist=str2similarity[row_distance_metric], labels=row_labels)
+    # # Set Sklearn's clustering model parameters
+    # row_model = AgglomerativeClustering(linkage=linkage_dic[clustering_method], n_clusters=2,
+    #                                     affinity=str2func[row_distance_metric])
+    #
+    # # fit Sklearn's clustering model
+    # row_model.fit(data)
+    # row_tree = make_tree(row_model, scipy=False)
+    # order_of_rows = order_leaves(row_model, tree=row_tree, data=data,
+    #                              dist=str2similarity[row_distance_metric], labels=row_labels)
+
+    #  # fastcluster
+    D = pdist(data, metric=cusca.custom_pearson_dist)
+    Z = fastcluster.linkage(D, method='weighted')
+    numeric_order_of_rows, R = two_plot_2_dendrogram(Z=Z, num_clust=2, no_plot=True)
+    # order_of_rows = [row_labels[int(i)] for i in numeric_order_of_rows]  # Getting label names from order of rows
+    # order_of_rows = row_labels[numeric_order_of_rows]  # Getting label names from order of rows
+    order_of_rows = [row_labels[i] for i in numeric_order_of_rows]
+
+    row_tree = make_tree(Z, scipy=True, n_leaves=len(order_of_rows))
+
     log("About to write gtr file", get_linenumber(), DEBUG)
     # Create gtr file
     make_gtr(row_tree, data=data, file_name=output_base_name+'.gtr', dist=str2similarity[row_distance_metric])

src/hc_functions.py

@@ -938,7 +938,7 @@ def parse_data(gct_name, row_normalization=False, col_normalization=False, row_c
 }
 
 
-def make_tree(model, data=None):
+def make_tree(model, n_leaves=None, scipy=False):
     """
     Modified from:
     https://stackoverflow.com/questions/27386641/how-to-traverse-a-tree-from-sklearn-agglomerativeclustering
@@ -967,7 +967,12 @@ def make_tree(model, data=None):
     # print(tree)
     # return tree
 
-    return dict(enumerate(model.children_, model.n_leaves_))
+    if scipy:
+        return dict(enumerate(np.delete(model, [2, 3], 1).astype(int), n_leaves))
+
+    else:
+
+        return dict(enumerate(model.children_, model.n_leaves_))
     # return dict(enumerate(model.children_, 1))
 
 
@@ -1249,3 +1254,14 @@ def better_dendodist(children, distance, tree, data, axis):
     for pair in children:
         distances_list.append(centroid_distances(pair[0], pair[1], tree, data, axis, distance=distance))
     return distances_list
+
+
+def two_plot_2_dendrogram(Z, num_clust, no_plot=True):
+    # plt.clf()
+    threshold = Z[-num_clust + 1, 2]
+    R = dendrogram(Z, no_labels=True, color_threshold=threshold, no_plot=no_plot)
+    order_of_columns = R['leaves']
+
+    # plt.show()
+    # plt.savefig('WHAT_ARE_THOOOOSE.png')
+    return order_of_columns, R