From 32c20f0a31ea5fc90d34bdda7d9bd370858db9ba Mon Sep 17 00:00:00 2001
From: barthes <nicolas.barthes@cefe.cnrs.fr>
Date: Tue, 30 Apr 2024 15:20:24 +0200
Subject: [PATCH] improved HDBSCAN clustering (random still buggy)
---
src/Class_Mod/HDBSCAN_Clustering.py | 13 ++++++++----
src/pages/1-samples_selection.py | 31 ++++++++++-------------------
2 files changed, 20 insertions(+), 24 deletions(-)
diff --git a/src/Class_Mod/HDBSCAN_Clustering.py b/src/Class_Mod/HDBSCAN_Clustering.py
index d19b990..d20aa7e 100644
--- a/src/Class_Mod/HDBSCAN_Clustering.py
+++ b/src/Class_Mod/HDBSCAN_Clustering.py
@@ -57,11 +57,15 @@ class Hdbscan:
self._ij_hdbscan_score = self.DBCV(self._clusterable_embedding, self._ij_label,)# dist_function=euclidean)
self._score.at[i,j] = self._ij_hdbscan_score
# get the best DBCV score
- self._hdbscan_score = max(self._score.max())
+ self._hdbscan_bscore = max(self._score.max())
# find the coordinates of the best clustering parameters and run HDBSCAN below
- self._bparams = np.where(self._score == self._hdbscan_score)
+ self._bparams = np.where(self._score == self._hdbscan_bscore)
# run HDBSCAN with best params
- self._labels = HDBSCAN(min_samples=self._param_dist['min_samples'][self._bparams[0][0]], min_cluster_size=self._param_dist['min_cluster_size'][self._bparams[1][0]], metric=self._param_dist['metric'][self._bparams[1][0]]).fit_predict(self._clusterable_embedding)
+
+ self.best_hdbscan = HDBSCAN(min_samples=self._param_dist['min_samples'][self._bparams[0][0]], min_cluster_size=self._param_dist['min_cluster_size'][self._bparams[1][0]], metric=self._param_dist['metric'][self._bparams[1][0]], store_centers="medoid", )
+ self.best_hdbscan.fit_predict(self._clusterable_embedding)
+ self._labels = self.best_hdbscan.labels_
+ self._centers = self.best_hdbscan.medoids_
def DBCV(self, X, labels, dist_function=euclidean):
"""
@@ -316,4 +320,5 @@ class Hdbscan:
@property
def HDBSCAN_scores_(self):
- return self._labels, self._hdbscan_score
+ clu = [f'cluster#{i}' for i in self._labels]
+ return clu, self._hdbscan_bscore, self._centers
diff --git a/src/pages/1-samples_selection.py b/src/pages/1-samples_selection.py
index de03b98..83c7c25 100644
--- a/src/pages/1-samples_selection.py
+++ b/src/pages/1-samples_selection.py
@@ -2,19 +2,8 @@ from Packages import *
st.set_page_config(page_title="NIRS Utils", page_icon=":goat:", layout="wide")
from Modules import *
-
-
-# HTML pour le bandeau "CEFE - CNRS"
-# bandeau_html = """
-# <div style="width: 100%; background-color: #4682B4; padding: 10px; margin-bottom: 10px;">
-# <h1 style="text-align: center; color: white;">CEFE - CNRS</h1>
-# </div>
-# """
-# # Injecter le code HTML du bandeau
-# st.markdown(bandeau_html, unsafe_allow_html=True)
add_header()
-
st.session_state["interface"] = st.session_state.get('interface')
if st.session_state["interface"] == 'simple':
hide_pages("Predictions")
@@ -152,10 +141,12 @@ if not t.empty:
data, labels, clu_centers = cl_model.fit_optimal(nclusters = ncluster)
elif clus_method == cluster_methods[2]:
- optimized_hdbscan = Hdbscan(np.array(t))
- labels, hdbscan_score = optimized_hdbscan.HDBSCAN_scores_
- non_clustered = np.where(labels == -1)
- labels[non_clustered] = 1000
+ optimized_hdbscan = Hdbscan(np.array(tcr))
+ labels, hdbscan_score, clu_centers = optimized_hdbscan.HDBSCAN_scores_
+ labels = np.array(labels)
+ non_clustered = np.where(np.array(labels) == "cluster#-1")
+ unclustered_number = len(non_clustered[0])
+ labels[non_clustered] = "unclust."
labels = labels.tolist()
elif clus_method == cluster_methods[3]:
@@ -193,7 +184,6 @@ if labels:
# list indexes of selected samples for colored plot
if selected_samples_idx:
-
if meta_data.empty:
sam = pd.DataFrame({'name': spectra.index[selected_samples_idx],
'cluster':np.array(labels)[selected_samples_idx]},
@@ -359,11 +349,12 @@ if not spectra.empty:
if dim_red_method == dim_red_methods[2] and clus_method == cluster_methods[2]: # UMAP clustered by HDBSCAN
with loadings: # Display some clustering metrics
st.write('Clustering metrics:')
- clusters_number = set(labels)
- clusters_number.remove(-1)
- st.write('Optimal number of clusters = ' + str(len(clusters_number)))
+ clusters_number = len(set(labels))
+ if unclustered_number > 0:
+ clusters_number -= 1
+ st.write('Optimal number of clusters = ' + str(clusters_number))
st.write('DBCV score (-1 to 1 - higher is better) = ' + str(round(hdbscan_score,3)))
- st.write('Unclassified samples: ' + str(len(t[labels==-1])) + ' on ' + str(len(t)) + ' samples (' + str(round(len(t[labels==-1])/len(t)*100, 1)) + '%).')
+ st.write('Unclassified samples: ' + str(unclustered_number) + ' on ' + str(len(tcr)) + ' samples (' + str(round(unclustered_number/len(tcr)*100, 1)) + '%).')
--
GitLab