File size: 7,805 Bytes
5e01175 ea572f9 5e01175 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 |
from typing import Literal, List, Tuple, Optional, Dict
from .protac_dataset import PROTAC_Dataset
import pandas as pd
from sklearn.base import ClassifierMixin
from sklearn.ensemble import (
RandomForestClassifier,
GradientBoostingClassifier,
)
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
import torch
import torch.nn as nn
from torchmetrics import (
Accuracy,
AUROC,
Precision,
Recall,
F1Score,
MetricCollection,
)
import optuna
def train_sklearn_model(
clf: ClassifierMixin,
protein2embedding: Dict,
cell2embedding: Dict,
smiles2fp: Dict,
train_df: pd.DataFrame,
val_df: pd.DataFrame,
test_df: Optional[pd.DataFrame] = None,
active_label: str = 'Active',
use_single_scaler: bool = True,
) -> Tuple[ClassifierMixin, Dict]:
""" Train a classifier model on train and val sets and evaluate it on a test set.
Args:
clf: The classifier model to train and evaluate.
train_df (pd.DataFrame): The training set.
val_df (pd.DataFrame): The validation set.
test_df (Optional[pd.DataFrame]): The test set.
Returns:
Tuple[ClassifierMixin, nn.ModuleDict]: The trained model and the metrics.
"""
# Initialize the datasets
train_ds = PROTAC_Dataset(
train_df,
protein2embedding,
cell2embedding,
smiles2fp,
active_label=active_label,
use_smote=False,
)
scaler = train_ds.fit_scaling(use_single_scaler=use_single_scaler)
train_ds.apply_scaling(scaler, use_single_scaler=use_single_scaler)
val_ds = PROTAC_Dataset(
val_df,
protein2embedding,
cell2embedding,
smiles2fp,
active_label=active_label,
use_smote=False,
)
val_ds.apply_scaling(scaler, use_single_scaler=use_single_scaler)
if test_df is not None:
test_ds = PROTAC_Dataset(
test_df,
protein2embedding,
cell2embedding,
smiles2fp,
active_label=active_label,
use_smote=False,
)
test_ds.apply_scaling(scaler, use_single_scaler=use_single_scaler)
# Get the numpy arrays
X_train, y_train = train_ds.get_numpy_arrays()
X_val, y_val = val_ds.get_numpy_arrays()
if test_df is not None:
X_test, y_test = test_ds.get_numpy_arrays()
# Train the model
clf.fit(X_train, y_train)
# Define the metrics as a module dict
stages = ['train_metrics', 'val_metrics', 'test_metrics']
metrics = nn.ModuleDict({s: MetricCollection({
'acc': Accuracy(task='binary'),
'roc_auc': AUROC(task='binary'),
'precision': Precision(task='binary'),
'recall': Recall(task='binary'),
'f1_score': F1Score(task='binary'),
'opt_score': Accuracy(task='binary') + F1Score(task='binary'),
'hp_metric': Accuracy(task='binary'),
}, prefix=s.replace('metrics', '')) for s in stages})
# Get the predictions
metrics_out = {}
y_pred = torch.tensor(clf.predict_proba(X_train)[:, 1])
y_true = torch.tensor(y_train)
metrics['train_metrics'].update(y_pred, y_true)
metrics_out.update(metrics['train_metrics'].compute())
y_pred = torch.tensor(clf.predict_proba(X_val)[:, 1])
y_true = torch.tensor(y_val)
metrics['val_metrics'].update(y_pred, y_true)
metrics_out.update(metrics['val_metrics'].compute())
if test_df is not None:
y_pred = torch.tensor(clf.predict_proba(X_test)[:, 1])
y_true = torch.tensor(y_test)
metrics['test_metrics'].update(y_pred, y_true)
metrics_out.update(metrics['test_metrics'].compute())
return clf, metrics_out
def suggest_random_forest(
trial: optuna.Trial,
) -> ClassifierMixin:
""" Suggest hyperparameters for a Random Forest classifier.
Args:
trial (optuna.Trial): The Optuna trial object.
Returns:
ClassifierMixin: The Random Forest classifier with the suggested hyperparameters.
"""
n_estimators = trial.suggest_int('model_n_estimators', 10, 1000)
max_depth = trial.suggest_int('model_max_depth', 2, 100)
min_samples_split = trial.suggest_int('model_min_samples_split', 2, 10)
min_samples_leaf = trial.suggest_int('model_min_samples_leaf', 1, 10)
max_features = trial.suggest_categorical('model_max_features', [None, 'sqrt', 'log2'])
criterion = trial.suggest_categorical('model_criterion', ['gini', 'entropy'])
clf = RandomForestClassifier(
n_estimators=n_estimators,
max_depth=max_depth,
min_samples_split=min_samples_split,
min_samples_leaf=min_samples_leaf,
max_features=max_features,
criterion=criterion,
random_state=42,
)
return clf
def suggest_logistic_regression(
trial: optuna.Trial,
) -> ClassifierMixin:
""" Suggest hyperparameters for a Logistic Regression classifier.
Args:
trial (optuna.Trial): The Optuna trial object.
Returns:
ClassifierMixin: The Logistic Regression classifier with the suggested hyperparameters.
"""
# Suggest values for the logistic regression hyperparameters
C = trial.suggest_loguniform('model_C', 1e-4, 1e2)
penalty = trial.suggest_categorical('model_penalty', ['l1', 'l2', 'elasticnet', None])
solver = trial.suggest_categorical('model_solver', ['newton-cholesky', 'lbfgs', 'liblinear', 'sag', 'saga'])
# Check solver compatibility
if penalty == 'l1' and solver not in ['liblinear', 'saga']:
raise optuna.exceptions.TrialPruned()
if penalty == None and solver not in ['newton-cholesky', 'lbfgs', 'sag']:
raise optuna.exceptions.TrialPruned()
# Configure the classifier with the trial's suggested parameters
clf = LogisticRegression(
C=C,
penalty=penalty,
solver=solver,
max_iter=1000,
random_state=42,
)
return clf
def suggest_svc(
trial: optuna.Trial,
) -> ClassifierMixin:
""" Suggest hyperparameters for an SVC classifier.
Args:
trial (optuna.Trial): The Optuna trial object.
Returns:
ClassifierMixin: The SVC classifier with the suggested hyperparameters.
"""
C = trial.suggest_loguniform('model_C', 1e-4, 1e2)
kernel = trial.suggest_categorical('model_kernel', ['linear', 'poly', 'rbf', 'sigmoid'])
gamma = trial.suggest_categorical('model_gamma', ['scale', 'auto'])
degree = trial.suggest_int('model_degree', 2, 5) if kernel == 'poly' else 3
clf = SVC(
C=C,
kernel=kernel,
gamma=gamma,
degree=degree,
probability=True,
random_state=42,
)
return clf
def suggest_gradient_boosting(
trial: optuna.Trial,
) -> ClassifierMixin:
""" Suggest hyperparameters for a Gradient Boosting classifier.
Args:
trial (optuna.Trial): The Optuna trial object.
Returns:
ClassifierMixin: The Gradient Boosting classifier with the suggested hyperparameters.
"""
n_estimators = trial.suggest_int('model_n_estimators', 50, 500)
learning_rate = trial.suggest_loguniform('model_learning_rate', 0.01, 1)
max_depth = trial.suggest_int('model_max_depth', 3, 10)
min_samples_split = trial.suggest_int('model_min_samples_split', 2, 10)
min_samples_leaf = trial.suggest_int('model_min_samples_leaf', 1, 10)
max_features = trial.suggest_categorical('model_max_features', ['sqrt', 'log2', None])
clf = GradientBoostingClassifier(
n_estimators=n_estimators,
learning_rate=learning_rate,
max_depth=max_depth,
min_samples_split=min_samples_split,
min_samples_leaf=min_samples_leaf,
max_features=max_features,
random_state=42,
)
return clf |