BaseEnvironment API
Overview
The BaseEnvironment class is the core abstract interface for MAT-HPO optimization. Inherit from this class to define your optimization problem.
️ Class Definition
python
from MAT_HPO_LIB import BaseEnvironment
class MyEnvironment(BaseEnvironment):
def __init__(self, name="MyEnvironment"):
super().__init__(name)
Required Methods
load_data()
load_data() -> Any
Load and preprocess your dataset. Store data as instance variables.
python
def load_data(self):
# Load your data
X, y = load_my_dataset()
# Store as instance variables
self.X_train, self.X_val = train_test_split(X, y)
return {"samples": len(X), "features": X.shape[1]}create_model(hyperparams)
create_model(hyperparams: dict) -> Any
Create a model with given hyperparameters.
python
def create_model(self, hyperparams):
model = MyModel(
hidden_size=hyperparams['hidden_size'],
learning_rate=hyperparams['learning_rate']
)
return modeltrain_evaluate(model, hyperparams)
train_evaluate(model: Any, hyperparams: dict) -> dict
Train the model and return evaluation metrics.
python
def train_evaluate(self, model, hyperparams):
# Train your model
model.fit(self.X_train, self.y_train)
# Evaluate performance
predictions = model.predict(self.X_val)
f1 = f1_score(self.y_val, predictions)
accuracy = accuracy_score(self.y_val, predictions)
return {'f1': f1, 'accuracy': accuracy}compute_reward(metrics)
compute_reward(metrics: dict) -> float
Compute reward from metrics (higher is better).
python
def compute_reward(self, metrics):
# Simple reward: optimize F1-score
return metrics['f1']
# Multi-objective reward
return 0.7 * metrics['f1'] + 0.3 * metrics['accuracy']Complete Example
python
from MAT_HPO_LIB import BaseEnvironment
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import f1_score, accuracy_score
class RandomForestEnvironment(BaseEnvironment):
def load_data(self):
# Load your dataset
X, y = self.load_my_data()
self.X_train, self.X_val, self.y_train, self.y_val = \
train_test_split(X, y, test_size=0.2, random_state=42)
return {"samples": len(X), "features": X.shape[1]}
def create_model(self, hyperparams):
return RandomForestClassifier(
n_estimators=int(hyperparams['n_estimators']),
max_depth=int(hyperparams['max_depth']),
random_state=42
)
def train_evaluate(self, model, hyperparams):
model.fit(self.X_train, self.y_train)
predictions = model.predict(self.X_val)
return {
'f1': f1_score(self.y_val, predictions, average='weighted'),
'accuracy': accuracy_score(self.y_val, predictions)
}
def compute_reward(self, metrics):
return metrics['f1'] # Optimize for F1-score