def fit(self, X, y, scoring="accuracy", crossval=5):
"""
Fit the Reinforcement classifier with data, either adding to previous previous data or learning for first time.
:param X: Input Features
:param y: Class Labels
:param scoring: Scoring used for cross validation
:param crossval: Cross Validation number of folds
:return: True if a new model is fit to the data, or a previous model is updated
False if old model when fit to new data performs poorly in comparison to
earlier data
"""
if not self.reTrain: # Train first time
score = cv.cross_val_score(self.clf, X, y, scoring, cv=crossval)
sqlManager.insertValue(self.name, 0.0, score.mean(), 0, len(y), 1) # Store the first result of clf
self.clf.fit(X, y)
joblib.dump(self.clf, "model.pkl") # Store the CLF
print("Data Fit")
return True
else:
previousData = sqlManager.selectNewestRecord(self.name) # Check the last entry of CLF
if len(previousData) > 0:
oldSize = previousData[5]
newSize = len(y)
accScore = previousData[3]
score = cv.cross_val_score(self.clf, X, y, scoring, cv=crossval)
newAccScore = score.mean()
print("Old Accuracy Score : ", accScore)
print("New Accuracy Score : ", newAccScore)
if accScore <= newAccScore: # If new data is benefitial, increases accuracy
print("Reinforcement Learning : Newer model is superior. Saving Model.")
self.clf.fit(X, y)
sqlManager.insertValue(self.name, accScore, newAccScore, oldSize, newSize, 1)
joblib.dump(self.clf, "model.pkl")
return True
else:
print("Reinforcement Learning : Newer model is inferior. Not saving model.")
return False
def __exit__(self, exc_type, exc_val, exc_tb):
sqlManager.close()
