fixes #12: force new model to be trained once data has been reloaded

mukeshmk · Jul 16, 2020 · 5165f6b · 5165f6b
1 parent 9ecec12
commit 5165f6b
Showing 1 changed file with 13 additions and 3 deletions.
diff --git a/src/fmlearn.py b/src/fmlearn.py
@@ -22,6 22,8 @@ def __init__(self):
  self._accuracy = None
  # to store feature encoders
  self._encoders = {}
  # boolean to track retraining of the model
  self._retain = False
 
  def get_encoders(self):
  return self._encoders
@@ -30,7 32,10 @@ def get_X_cols(self):
  return self._X.columns
 
  def load_data(self):
- # TODO: force new model to be trained once data has been reloaded?
  # force new model to be trained once data has been reloaded
  # retraining of model occurs only when the predict() function is being called
  # this is due to absense of background processes framework in the application.
  self._retain = True
 
  # loads data from the SQL database and pre-processes the data.
  self._df = utils.get_df_from_db()
@@ -68,12 73,17 @@ def load_data_and_train(self):
  self.train()
 
  def predict(self, X_pred):
- # TODO: force retrain of model if the model is older than a set time frame?
-
  # check if the shape of the input df matches that used to train the model.
  if X_pred.shape[1] != self._X.shape[1]:
  raise RuntimeError('Input Shape miss match! aborting!')
 
  # TODO: force retrain of model if the model is older than a set time frame?
  # or if a set of new data records have been added to the model.
  # at this point reload the data and train the model.
  if self._retain == True:
  self.train()
  self._retain = False
 
  y_pred = self._model.predict(X_pred)
 
  # decodes the predicted value using the inverse_transform method of the encoder.