From 4b97a50f8cfa7540d20b1966daedefbdb8876d84 Mon Sep 17 00:00:00 2001
From: Nicolas Barthes <nicolas.barthes@cnrs.fr>
Date: Fri, 12 Apr 2024 16:03:23 +0200
Subject: [PATCH] updated HDBSCAN optimization
---
Class_Mod/HDBSCAN_Clustering.py | 26 ++++++++++++++++++++------
Packages.py | 2 +-
pages/1-samples_selection.py | 13 +++++--------
3 files changed, 26 insertions(+), 15 deletions(-)
diff --git a/Class_Mod/HDBSCAN_Clustering.py b/Class_Mod/HDBSCAN_Clustering.py
index 0d9268b..f019282 100644
--- a/Class_Mod/HDBSCAN_Clustering.py
+++ b/Class_Mod/HDBSCAN_Clustering.py
@@ -256,14 +256,18 @@ def _get_label_members(X, labels, cluster):
members = X[indices]
return members
-def HDBSCAN_function(data, min_cluster_size):
+def HDBSCAN_function(data):
# param_dist = {'min_samples': [1,5,10,30],
# 'min_cluster_size':[5,10,20,30,50,75,100],
# # 'cluster_selection_method' : ['eom','leaf'],
# # 'metric' : ['euclidean','manhattan']
# }
- param_dist = {'min_samples': [1,5],
- 'min_cluster_size':[5,10],
+ # param_dist = {'min_samples': [1,5,10,50],
+ # 'min_cluster_size':[5,10,30,50,100,300,500],
+ # }
+ param_dist = {'min_samples': [1,5, 10,],
+ 'min_cluster_size':[5,10,30,50,100],
+ 'metric' : ['euclidean','manhattan'],
}
clusterable_embedding = UMAP(
@@ -272,6 +276,15 @@ def HDBSCAN_function(data, min_cluster_size):
n_components=5,
random_state=42,
).fit_transform(data)
+
+ # RandomizedSearchCV not working...
+ # def scoring(model, clusterable_embedding):
+ # label = HDBSCAN().fit_predict(clusterable_embedding)
+ # hdbscan_score = DBCV(clusterable_embedding, label, dist_function=euclidean)
+ # return hdbscan_score
+ # tunning = RandomizedSearchCV(estimator=HDBSCAN(), param_distributions=param_dist, scoring=scoring)
+ # tunning.fit(clusterable_embedding)
+ # return tunning
min_score = pd.DataFrame()
for i in param_dist.get('min_samples'):
for j in param_dist.get('min_cluster_size'):
@@ -279,7 +292,8 @@ def HDBSCAN_function(data, min_cluster_size):
ij_hdbscan_score = DBCV(clusterable_embedding, ij_label, dist_function=euclidean)
min_score.at[i,j] = ij_hdbscan_score
hdbscan_score = max(min_score.max())
- # get the coordinates of the best clustering paramters and run HDBSCAN below
-
- labels = HDBSCAN(min_samples=1, min_cluster_size=min_cluster_size).fit_predict(clusterable_embedding)
+ # get the coordinates of the best clustering parameters and run HDBSCAN below
+ bparams = np.where(min_score == hdbscan_score)
+ # run HDBSCAN with best params
+ labels = HDBSCAN(min_samples=param_dist['min_samples'][bparams[0][0]], min_cluster_size=param_dist['min_cluster_size'][bparams[1][0]], metric=param_dist['metric'][bparams[1][0]]).fit_predict(clusterable_embedding)
return labels, hdbscan_score
diff --git a/Packages.py b/Packages.py
index 217daba..9cad07b 100644
--- a/Packages.py
+++ b/Packages.py
@@ -27,7 +27,7 @@ from pinard import utils
from pinard import preprocessing as pp
from pinard.model_selection import train_test_split_idx
-from sklearn.model_selection import train_test_split, cross_val_score, cross_val_predict, cross_validate
+from sklearn.model_selection import train_test_split, cross_val_score, cross_val_predict, cross_validate, RandomizedSearchCV
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.compose import TransformedTargetRegressor
from sklearn.metrics import mean_absolute_error, mean_squared_error, mean_absolute_percentage_error, r2_score
diff --git a/pages/1-samples_selection.py b/pages/1-samples_selection.py
index 579b528..cb1348d 100644
--- a/pages/1-samples_selection.py
+++ b/pages/1-samples_selection.py
@@ -119,8 +119,7 @@ with container2:
elif type_cluster == 'HDBSCAN':
from Class_Mod.HDBSCAN_Clustering import HDBSCAN_function
- labels, hdbscan_score = HDBSCAN_function(data_import, min_cluster_size=10)
-
+ labels, hdbscan_score = HDBSCAN_function(data_import)
with scores:
t = model.scores_
if type_cluster in ['AP', 'Kmeans']:
@@ -138,12 +137,10 @@ with container2:
elif type_cluster in ['HDBSCAN']:
st.write('plot HDBSCAN clustering')
- fig_hdbscan = px.scatter_3d(t, x=axis1, y=axis2, z = axis3, color=labels)
- fig_hdbscan.update_traces(marker=dict(size=4))
- st.plotly_chart(fig_hdbscan)
- st.write('DBCV score = ' + str(hdbscan_score))
- # st.dataframe(min_score.stack().agg(['min']))
-
+ fig = px.scatter_3d(t, x=axis1, y=axis2, z = axis3, color=labels)
+ fig.update_traces(marker=dict(size=4))
+ # st.plotly_chart(fig_hdbscan)
+ st.write('DBCV score (-1:1) = ' + str(hdbscan_score))
else:
if test == '.dx':
--
GitLab