Module reclab.recommenders.llorma.llorma
Tensorflow implementation of AutoRec recommender.
Expand source code
"""Tensorflow implementation of AutoRec recommender."""
import numpy as np
from .llorma_lib import llorma_g
from .. import recommender
class Llorma(recommender.PredictRecommender):
"""Many local low rank models averaged via kernels.
Parameters
----------
max_user : int
Maximum number of users in the environment
max_item : int
Maximum number of items in the environment
n_anchor : int
Number of local model to build in the train phase
pre_rank : int
Dimension of the pre-train user/item latent factors
rank : int
Dimension of the train user/item factors
pre_lambda_val : float
Regularization parameter for the pre-train matrix factorization
lambda_val : float
Regularization parameter for the train model
pre_learning_rate : float
Learning rate when optimizing the pre-train matrix factorization
learning_rate : float
Learning rate for the the train model
pre_train_steps : int
Number of epochs in the pre-train phase
train_steps : int
Number of epochs in the training phase
batch_size : int
Batch size in training phase
use_cache : bool
If True use stored pre-trained item/user latent factors
results_path :
Folder to save model outputs and checkpoints.
kernel_fun : callable
kernel function used for similarity,
"""
def __init__(self,
max_user,
max_item,
n_anchor=10,
pre_rank=5,
pre_learning_rate=1e-3,
pre_lambda_val=10,
pre_train_steps=10,
rank=10,
learning_rate=1e-2,
lambda_val=1e-3,
train_steps=10,
batch_size=128,
use_cache=True,
result_path='results',
kernel_fun=None):
"""Create new Local Low-Rank Matrix Approximation (LLORMA) recommender."""
super().__init__()
self.model = llorma_g.Llorma(max_user, max_item, n_anchor, pre_rank,
pre_learning_rate, pre_lambda_val, pre_train_steps,
rank, learning_rate, lambda_val, train_steps,
batch_size, use_cache, result_path, kernel_fun)
self._hyperparameters.update(locals())
# We only want the function arguments so remove class related objects.
del self._hyperparameters['self']
del self._hyperparameters['__class__']
@property
def name(self): # noqa: D102
return 'llorma'
def _predict(self, user_item): # noqa: D102
users, items, _ = list(zip(*user_item))
users = np.array(users)
items = np.array(items)
# Check that both the item and the user have been seen in historical data.
is_seen_uid = np.array(users <= (self.model.batch_manager.n_user - 1))
is_seen_iid = np.array(items <= (self.model.batch_manager.n_item - 1))
is_seen_id = np.logical_and(is_seen_iid, is_seen_uid)
seen_user_item = np.column_stack((users[is_seen_id], items[is_seen_id]))
seen_estimate = self.model.predict(seen_user_item)
# Choose the mean of the seen values as the estimate for the unseen ids.
unseen_estimate = np.mean(seen_estimate)
estimate = np.ones(len(users))*unseen_estimate
estimate[is_seen_id] = seen_estimate
print('Low: {:.3f}, Mean: {:.3f}, High: {:.3f}'.format(np.quantile(seen_estimate, 0.25),
np.quantile(seen_estimate, 0.5),
np.quantile(seen_estimate, 0.75)))
return estimate
def update(self, users=None, items=None, ratings=None): # noqa: D102
super().update(users, items, ratings)
updated_ratings = dict(self._ratings)
user_items = np.array(list(updated_ratings.keys()))
rating_arr = list(updated_ratings.values())
data = np.column_stack((user_items, rating_arr))
self.model.reset_data(data, data, data)
self.model.train()Classes
class Llorma (max_user, max_item, n_anchor=10, pre_rank=5, pre_learning_rate=0.001, pre_lambda_val=10, pre_train_steps=10, rank=10, learning_rate=0.01, lambda_val=0.001, train_steps=10, batch_size=128, use_cache=True, result_path='results', kernel_fun=None)-
Many local low rank models averaged via kernels.
Parameters
max_user:int- Maximum number of users in the environment
- max_item : int
- Maximum number of items in the environment
n_anchor:int- Number of local model to build in the train phase
pre_rank:int- Dimension of the pre-train user/item latent factors
rank:int- Dimension of the train user/item factors
pre_lambda_val:float- Regularization parameter for the pre-train matrix factorization
lambda_val:float- Regularization parameter for the train model
pre_learning_rate:float- Learning rate when optimizing the pre-train matrix factorization
learning_rate:float- Learning rate for the the train model
pre_train_steps:int- Number of epochs in the pre-train phase
train_steps:int- Number of epochs in the training phase
batch_size:int- Batch size in training phase
use_cache:bool- If True use stored pre-trained item/user latent factors
- results_path :
- Folder to save model outputs and checkpoints.
kernel_fun:callable- kernel function used for similarity,
Create new Local Low-Rank Matrix Approximation (LLORMA) recommender.
Expand source code
class Llorma(recommender.PredictRecommender): """Many local low rank models averaged via kernels. Parameters ---------- max_user : int Maximum number of users in the environment max_item : int Maximum number of items in the environment n_anchor : int Number of local model to build in the train phase pre_rank : int Dimension of the pre-train user/item latent factors rank : int Dimension of the train user/item factors pre_lambda_val : float Regularization parameter for the pre-train matrix factorization lambda_val : float Regularization parameter for the train model pre_learning_rate : float Learning rate when optimizing the pre-train matrix factorization learning_rate : float Learning rate for the the train model pre_train_steps : int Number of epochs in the pre-train phase train_steps : int Number of epochs in the training phase batch_size : int Batch size in training phase use_cache : bool If True use stored pre-trained item/user latent factors results_path : Folder to save model outputs and checkpoints. kernel_fun : callable kernel function used for similarity, """ def __init__(self, max_user, max_item, n_anchor=10, pre_rank=5, pre_learning_rate=1e-3, pre_lambda_val=10, pre_train_steps=10, rank=10, learning_rate=1e-2, lambda_val=1e-3, train_steps=10, batch_size=128, use_cache=True, result_path='results', kernel_fun=None): """Create new Local Low-Rank Matrix Approximation (LLORMA) recommender.""" super().__init__() self.model = llorma_g.Llorma(max_user, max_item, n_anchor, pre_rank, pre_learning_rate, pre_lambda_val, pre_train_steps, rank, learning_rate, lambda_val, train_steps, batch_size, use_cache, result_path, kernel_fun) self._hyperparameters.update(locals()) # We only want the function arguments so remove class related objects. del self._hyperparameters['self'] del self._hyperparameters['__class__'] @property def name(self): # noqa: D102 return 'llorma' def _predict(self, user_item): # noqa: D102 users, items, _ = list(zip(*user_item)) users = np.array(users) items = np.array(items) # Check that both the item and the user have been seen in historical data. is_seen_uid = np.array(users <= (self.model.batch_manager.n_user - 1)) is_seen_iid = np.array(items <= (self.model.batch_manager.n_item - 1)) is_seen_id = np.logical_and(is_seen_iid, is_seen_uid) seen_user_item = np.column_stack((users[is_seen_id], items[is_seen_id])) seen_estimate = self.model.predict(seen_user_item) # Choose the mean of the seen values as the estimate for the unseen ids. unseen_estimate = np.mean(seen_estimate) estimate = np.ones(len(users))*unseen_estimate estimate[is_seen_id] = seen_estimate print('Low: {:.3f}, Mean: {:.3f}, High: {:.3f}'.format(np.quantile(seen_estimate, 0.25), np.quantile(seen_estimate, 0.5), np.quantile(seen_estimate, 0.75))) return estimate def update(self, users=None, items=None, ratings=None): # noqa: D102 super().update(users, items, ratings) updated_ratings = dict(self._ratings) user_items = np.array(list(updated_ratings.keys())) rating_arr = list(updated_ratings.values()) data = np.column_stack((user_items, rating_arr)) self.model.reset_data(data, data, data) self.model.train()Ancestors
- PredictRecommender
- Recommender
- abc.ABC
Inherited members