# Copyright 2018 The Cornac Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
import numpy as np
from tqdm.auto import trange
from ..recommender import Recommender
from ..recommender import ANNMixin, MEASURE_DOT
from ...exception import ScoreException
from ...utils import get_rng
from ...utils.init_utils import xavier_uniform
[docs]
class CVAE(Recommender):
"""
Collaborative Variational Autoencoder
Parameters
----------
z_dim: int, optional, default: 50
The dimension of the user and item latent factors.
n_epochs: int, optional, default: 100
Maximum number of epochs for training.
lambda_u: float, optional, default: 1e-4
The regularization hyper-parameter for user latent factor.
lambda_v: float, optional, default: 0.001
The regularization hyper-parameter for item latent factor.
lambda_r: float, optional, default: 10.0
Parameter that balance the focus on content or ratings
lambda_w: float, optional, default: 1e-4
The regularization for VAE weights
lr: float, optional, default: 0.001
Learning rate in the auto-encoder training
a: float, optional, default: 1
The confidence of observed ratings.
b: float, optional, default: 0.01
The confidence of unseen ratings.
input_dim: int, optional, default: 8000
The size of input vector
vae_layers: list, optional, default: [200, 100]
The list containing size of each layers in neural network structure
act_fn: str, default: 'sigmoid'
Name of the activation function used for the variational auto-encoder.
Supported functions: ['sigmoid', 'tanh', 'elu', 'relu', 'relu6', 'leaky_relu', 'identity']
loss_type: String, optional, default: "cross-entropy"
Either "cross-entropy" or "rmse"
The type of loss function in the last layer
batch_size: int, optional, default: 128
The batch size for SGD.
init_params: dict, optional, default: {'U':None, 'V':None}
Initial U and V latent matrix
trainable: boolean, optional, default: True
When False, the model is not trained and Cornac assumes that the model already \
pre-trained (U and V are not None).
References
----------
Collaborative Variational Autoencoder for Recommender Systems
X. Li and J. She ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 2017
http://eelxpeng.github.io/assets/paper/Collaborative_Variational_Autoencoder.pdf
"""
def __init__(
self,
name="CVAE",
z_dim=50,
n_epochs=100,
lambda_u=1e-4,
lambda_v=0.001,
lambda_r=10,
lambda_w=1e-4,
lr=0.001,
a=1,
b=0.01,
input_dim=8000,
vae_layers=[200, 100],
act_fn="sigmoid",
loss_type="cross-entropy",
batch_size=128,
init_params=None,
trainable=True,
seed=None,
verbose=True,
):
super().__init__(name=name, trainable=trainable, verbose=verbose)
self.lambda_u = lambda_u
self.lambda_v = lambda_v
self.lambda_r = lambda_r
self.lambda_w = lambda_w
self.a = a
self.b = b
self.n_epochs = n_epochs
self.input_dim = input_dim
self.vae_layers = vae_layers
self.z_dim = z_dim
self.loss_type = loss_type
self.act_fn = act_fn
self.lr = lr
self.batch_size = batch_size
self.seed = seed
# Init params if provided
self.init_params = {} if init_params is None else init_params
self.U = self.init_params.get("U", None)
self.V = self.init_params.get("V", None)
def _init(self):
rng = get_rng(self.seed)
if self.U is None:
self.U = xavier_uniform((self.num_users, self.z_dim), rng)
if self.V is None:
self.V = xavier_uniform((self.num_items, self.z_dim), rng)
[docs]
def fit(self, train_set, val_set=None):
"""Fit the model to observations.
Parameters
----------
train_set: :obj:`cornac.data.Dataset`, required
User-Item preference data as well as additional modalities.
val_set: :obj:`cornac.data.Dataset`, optional, default: None
User-Item preference data for model selection purposes (e.g., early stopping).
Returns
-------
self : object
"""
Recommender.fit(self, train_set, val_set)
self._init()
if self.trainable:
self._fit_cvae(train_set)
return self
def _fit_cvae(self, train_set):
R = train_set.csc_matrix # csc for efficient slicing over items
document = train_set.item_text.batch_bow(
np.arange(train_set.num_items)
) # bag-of-words features
document = (document - document.min()) / (
document.max() - document.min()
) # normalization
# VAE initialization
from .cvae import Model
import tensorflow.compat.v1 as tf
tf.disable_eager_execution()
tf.set_random_seed(self.seed)
model = Model(
n_users=train_set.num_users,
n_items=train_set.num_items,
input_dim=self.input_dim,
U=self.U,
V=self.V,
n_z=self.z_dim,
lambda_u=self.lambda_u,
lambda_v=self.lambda_v,
lambda_r=self.lambda_r,
lambda_w=self.lambda_w,
layers=self.vae_layers,
loss_type=self.loss_type,
act_fn=self.act_fn,
seed=self.seed,
lr=self.lr,
)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer()) # init variable
loop = trange(self.n_epochs, disable=not self.verbose)
for _ in loop:
cf_loss, vae_loss, count = 0, 0, 0
for i, batch_ids in enumerate(
train_set.item_iter(self.batch_size, shuffle=True)
):
batch_R = R[:, batch_ids]
batch_C = np.ones(batch_R.shape) * self.b
batch_C[batch_R.nonzero()] = self.a
feed_dict = {
model.x: document[batch_ids],
model.ratings: batch_R.A,
model.C: batch_C,
model.item_ids: batch_ids,
}
_, _vae_los = sess.run([model.vae_update, model.vae_loss], feed_dict)
_, _cf_loss = sess.run([model.cf_update, model.cf_loss], feed_dict)
cf_loss += _cf_loss
vae_loss += _vae_los
count += len(batch_ids)
if i % 10 == 0:
loop.set_postfix(
vae_loss=(vae_loss / count), cf_loss=(cf_loss / count)
)
self.U, self.V = sess.run([model.U, model.V])
tf.reset_default_graph()
[docs]
def score(self, user_idx, item_idx=None):
"""Predict the scores/ratings of a user for an item.
Parameters
----------
user_idx: int, required
The index of the user for whom to perform score prediction.
item_idx: int, optional, default: None
The index of the item for which to perform score prediction.
If None, scores for all known items will be returned.
Returns
-------
res : A scalar or a Numpy array
Relative scores that the user gives to the item or to all known items
"""
if self.is_unknown_user(user_idx):
raise ScoreException("Can't make score prediction for user %d" % user_idx)
if item_idx is not None and self.is_unknown_item(item_idx):
raise ScoreException("Can't make score prediction for item %d" % item_idx)
if item_idx is None:
return self.V.dot(self.U[user_idx, :])
return self.V[item_idx, :].dot(self.U[user_idx, :])
[docs]
def get_vector_measure(self):
"""Getting a valid choice of vector measurement in ANNMixin._measures.
Returns
-------
measure: MEASURE_DOT
Dot product aka. inner product
"""
return MEASURE_DOT
[docs]
def get_user_vectors(self):
"""Getting a matrix of user vectors serving as query for ANN search.
Returns
-------
out: numpy.array
Matrix of user vectors for all users available in the model.
"""
return self.U
[docs]
def get_item_vectors(self):
"""Getting a matrix of item vectors used for building the index for ANN search.
Returns
-------
out: numpy.array
Matrix of item vectors for all items available in the model.
"""
return self.V