Merge pull request #138 from daisybio/development

New version: adapted hyperparameter files, faster GradientBoosting

Author: PascalIversen

docs/conf.py

@@ -56,9 +56,9 @@
 # the built documents.
 #
 # The short X.Y version.
-version = "1.2.0"
+version = "1.2.1"
 # The full version, including alpha/beta/rc tags.
-release = "1.2.0"
+release = "1.2.1"
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

drevalpy/models/DIPK/hyperparameters.yaml

@@ -3,7 +3,7 @@ DIPK:
   batch_size:
     - 64
   lr:
-    - 0.00001
+    - 0.0001
   heads:
     - 2
   fc_layer_num:

drevalpy/models/MOLIR/hyperparameters.yaml

@@ -4,15 +4,12 @@ MOLIR:
     - 32
   h_dim1:
     - 64
-    - 32
     - 16
   h_dim2:
     - 64
-    - 32
     - 16
   h_dim3:
     - 64
-    - 32
     - 16
   learning_rate:
     - 0.01

drevalpy/models/SimpleNeuralNetwork/hyperparameters.yaml

@@ -1,13 +1,8 @@
 ---
 SimpleNeuralNetwork:
   dropout_prob:
-    - 0.2
     - 0.3
-    - 0.4
   units_per_layer:
-    - - 10
-      - 10
-      - 10
     - - 32
       - 16
       - 8
@@ -27,9 +22,7 @@ SimpleNeuralNetwork:
 
 MultiOmicsNeuralNetwork:
   dropout_prob:
-    - 0.2
     - 0.3
-    - 0.4
   units_per_layer:
     - - 16
       - 8

drevalpy/models/SuperFELTR/hyperparameters.yaml

@@ -2,7 +2,6 @@
 SuperFELTR:
   mini_batch: 55
   dropout_rate:
-    - 0.3
     - 0.5
   weight_decay: 0.01
   out_dim_expr_encoder: 256
@@ -13,13 +12,22 @@ SuperFELTR:
     GDSC1: 0.1
     GDSC2: 0.1
     Toy_Data: 0.03
+    CCLE: 0.1
+    CTRPv1: 0.1
+    CTRPv2: 0.1
   mutation_var_threshold:
     GDSC1: 0.1
     GDSC2: 0.1
     Toy_Data: 0.05
+    CCLE: 0.1
+    CTRPv1: 0.1
+    CTRPv2: 0.1
   cnv_var_threshold:
     GDSC1: 0.7
     GDSC2: 0.7
     Toy_Data: 0.6
+    CCLE: 0.7
+    CTRPv1: 0.7
+    CTRPv2: 0.7
   margin: 1.0
   learning_rate: 0.01

drevalpy/models/baselines/hyperparameters.yaml

@@ -21,34 +21,28 @@ RandomForest:
   n_estimators:
     - 100
   max_depth:
-    - None
     - 5
     - 10
     - 30
-    - 50
   max_samples:
-    - 0.5
+    - 0.2
   n_jobs:
     - -1
   criterion:
     - squared_error
-    - absolute_error
 MultiOmicsRandomForest:
   n_estimators:
     - 100
   max_depth:
-    - None
     - 5
     - 10
     - 30
-    - 50
   max_samples:
-    - 0.5
+    - 0.2
   n_jobs:
     - -1
   criterion:
     - squared_error
-    - absolute_error
   n_components:
     - 100
 SVR:
@@ -73,34 +67,25 @@ SingleDrugRandomForest:
   n_estimators:
     - 100
   max_depth:
-    - None
     - 5
     - 10
     - 30
-    - 50
   max_samples:
-    - 0.5
+    - 0.2
   n_jobs:
     - -1
   criterion:
     - squared_error
-    - absolute_error
 GradientBoosting:
-  n_estimators:
+  max_iter:
     - 100
   learning_rate:
     - 0.1
     - 0.01
   max_depth:
-    - None
     - 5
     - 10
     - 30
-    - 50
-  subsample:
-    - 1.0
-    - 0.8
-    - 0.5
 SingleDrugElasticNet:
   l1_ratio:
     - 0.2

drevalpy/models/baselines/sklearn_models.py

@@ -1,7 +1,7 @@
 """Contains sklearn baseline models: ElasticNet, RandomForest, SVM."""
 
 import numpy as np
-from sklearn.ensemble import GradientBoostingRegressor, RandomForestRegressor
+from sklearn.ensemble import HistGradientBoostingRegressor, RandomForestRegressor
 from sklearn.linear_model import ElasticNet, Lasso, Ridge
 from sklearn.svm import SVR
 
@@ -237,9 +237,8 @@ def build_model(self, hyperparameters: dict):
         """
         if hyperparameters["max_depth"] == "None":
             hyperparameters["max_depth"] = None
-        self.model = GradientBoostingRegressor(
-            n_estimators=hyperparameters.get("n_estimators", 100),
+        self.model = HistGradientBoostingRegressor(
+            max_iter=hyperparameters.get("max_iter", 100),
             learning_rate=hyperparameters.get("learning_rate", 0.1),
             max_depth=hyperparameters.get("max_depth", 3),
-            subsample=hyperparameters.get("subsample", 1.0),
         )

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "drevalpy"
-version = "1.2.0"
+version = "1.2.1"
 description = "Drug response evaluation of cancer cell line drug response models in a fair setting"
 authors = ["DrEvalPy development team"]
 license = "GPL-3.0"