def movie_median_deviation_user_rescaling(df_train, df_test):
rescaler = Rescaler(df_train)
df_train_normalized = rescaler.normalize_deviation()
prediction_normalized = movie_median_deviation_user(df_train_normalized, df_test)
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def predictions_ALS_rescaled(train, test, **kwargs):
"""
ALS with PySpark rescaled.
First, a rescaling of the user such that they all have the same average of rating is done.
Then, the predictions are done using the function prediction_ALS().
Finally, the predictions are rescaled to recover the deviation of each user.
Args:
train (pandas.DataFrame): train set
test (pandas.DataFrame): test set
**kwargs: Arbitrary keyword arguments. Directly given to predictions_ALS().
Returns:
pandas.DataFrame: predictions, sorted by (Movie, User)
"""
# Load the class Rescaler
rescaler = Rescaler(train)
# Normalize the train data
df_train_normalized = rescaler.normalize_deviation()
# Predict using the normalized trained data
prediction_normalized = predictions_ALS(df_train_normalized, test, **kwargs)
# Rescale the prediction to recover the deviations
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def svd_rescaled(train, test, **kwargs):
"""
Singular Value Decomposition from library Surprise rescaled
(Based on Matrix Factorization)
First, a rescaling of the user such that they all have the same average of rating is done.
Then, the predictions are done using the function svd().
Finally, the predictions are rescaled to recover the deviation of each user.
Args:
train (pandas.DataFrame): train set
test (pandas.DataFrame): test set
**kwargs: Arbitrary keyword arguments. Directly given to svd().
Returns:
pandas.DataFrame: predictions, sorted by (Movie, User)
"""
# Load the class Rescaler
rescaler = Rescaler(train)
# Normalize the train data
df_train_normalized = rescaler.normalize_deviation()
# Predict using the normalized trained data
prediction_normalized = svd(df_train_normalized, test, **kwargs)
# Rescale the prediction to recover the deviations
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def baseline_rescaled(train, test):
"""
BaselineOnly from library Surprise rescaled
First, a rescaling of the user such that they all have the same average of rating is done.
Then, the predictions are done using the function baseline().
Finally, the predictions are rescaled to recover the deviation of each user.
Args:
train (pandas.DataFrame): train set
test (pandas.DataFrame): test set
Returns:
pandas.DataFrame: predictions, sorted by (Movie, User)
"""
# Load the class Rescaler
rescaler = Rescaler(train)
# Normalize the train data
df_train_normalized = rescaler.normalize_deviation()
# Predict using the normalized trained data
prediction_normalized = baseline(df_train_normalized, test)
# Rescale the prediction to recover the deviations
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def movie_median_deviation_user_rescaled(train, test):
"""
Movie median rescaled with the 'deviation_per_user' file and rescaled again.
First, a rescaling of the user such that they all have the same average of rating is done.
Then, the predictions are done using the function movie_median_deviation_user().
Finally, the predictions are rescaled to recover the deviation of each user.
Args:
train (pandas.DataFrame): train set
test (pandas.DataFrame): test set
Returns:
pandas.DataFrame: predictions, sorted by (Movie, User)
"""
# Load the class Rescaler
rescaler = Rescaler(train)
# Normalize the train data
df_train_normalized = rescaler.normalize_deviation()
# Predict using the normalized trained data
prediction_normalized = movie_median_deviation_user(
df_train_normalized, test)
# Rescale the prediction to recover the deviations
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def svd_rescaling(df_train, df_test, **kwargs):
rescaler = Rescaler(df_train)
df_train_normalized = rescaler.normalize_deviation()
prediction_normalized = svd(df_train_normalized, df_test, **kwargs)
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def baseline_rescaling(df_train, df_test):
rescaler = Rescaler(df_train)
df_train_normalized = rescaler.normalize_deviation()
prediction_normalized = baseline(df_train_normalized, df_test)
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def slope_one(df_train, df_test):
rescaler = Rescaler(df_train)
df_train_normalized = rescaler.normalize_deviation()
prediction_normalized = slope_one(df_train_normalized, df_test)
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction
def pyfm_rescaling(df_train, df_test, **kwargs):
"""
pyFM
First do a rescaling of the user in a way that they all have the same mean of rating.
This counter the effect of "mood" of users. Some of them given worst/better grade even if they have the same
appreciation of a movie.
:param df_train:
:param df_test:
:param kwargs:
gamma (float): regularization parameter
n_features (int): number of features for matrices
n_iter (int): number of iterations
init_method ('global_mean' or 'movie_mean'): kind of initial matrices (better result with 'global_mean')
:return:
"""
rescaler = Rescaler(df_train)
df_train_normalized = rescaler.normalize_deviation()
prediction_normalized = pyfm(df_train_normalized, df_test, **kwargs)
prediction = rescaler.recover_deviation(prediction_normalized)
return prediction