# -*- coding: utf-8 -*-
#
# Copyright 2020 Data61, CSIRO
#
# 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 tensorflow as tf
from tensorflow.keras import backend as K
import numpy as np
from ..core import StellarGraph
from ..core.validation import require_integer_in_range
from ..core.utils import normalize_adj
[docs]class AdjacencyPowerGenerator:
"""
A data generator for use with the Watch Your Step algorithm [1]. It calculates and returns the first `num_powers`
of the adjacency matrix row by row.
Args:
G (StellarGraph): a machine-learning StellarGraph-type graph
num_powers (int): the number of adjacency powers to calculate. Defaults
to `10` as this value was found to perform well by the authors of the paper.
"""
def __init__(self, G, num_powers=10):
if not isinstance(G, StellarGraph):
raise TypeError("G must be a StellarGraph object.")
require_integer_in_range(num_powers, "num_powers", min_val=1)
Aadj = G.to_adjacency_matrix().tocoo()
indices = np.column_stack((Aadj.col, Aadj.row))
self.Aadj_T = tf.sparse.SparseTensor(
indices=indices,
values=Aadj.data.astype(np.float32),
dense_shape=Aadj.shape,
)
self.transition_matrix_T = tf.sparse.SparseTensor(
indices=indices,
values=normalize_adj(Aadj, symmetric=False).data.astype(np.float32),
dense_shape=Aadj.shape,
)
self.num_powers = num_powers
[docs] def flow(self, batch_size, num_parallel_calls=1):
"""
Creates the `tensorflow.data.Dataset` object for training node embeddings from powers of the adjacency matrix.
Args:
batch_size (int): the number of rows of the adjacency powers to include in each batch.
num_parallel_calls (int): the number of threads to use for pre-processing of batches.
Returns:
A `tensorflow.data.Dataset` object for training node embeddings from powers of the adjacency matrix.
"""
require_integer_in_range(batch_size, "batch_size", min_val=1)
require_integer_in_range(num_parallel_calls, "num_parallel_calls", min_val=1)
row_dataset = tf.data.Dataset.from_tensor_slices(
tf.sparse.eye(int(self.Aadj_T.shape[0]))
)
adj_powers_dataset = row_dataset.map(
lambda ohe_rows: _partial_powers(
ohe_rows, self.transition_matrix_T, num_powers=self.num_powers
),
num_parallel_calls=num_parallel_calls,
)
row_index_dataset = tf.data.Dataset.range(self.Aadj_T.shape[0])
row_index_adj_powers_dataset = tf.data.Dataset.zip(
(row_index_dataset, adj_powers_dataset)
)
batch_adj_dataset = row_dataset.map(
lambda ohe_rows: _select_row_from_sparse_tensor(ohe_rows, self.Aadj_T),
num_parallel_calls=num_parallel_calls,
)
training_dataset = tf.data.Dataset.zip(
(row_index_adj_powers_dataset, batch_adj_dataset)
).batch(batch_size)
return training_dataset.repeat()
def _partial_powers(one_hot_encoded_row, Aadj_T, num_powers):
"""
This function computes the first num_powers powers of the adjacency matrix
for the row specified in one_hot_encoded_row
Args:
one_hot_encoded_row: one-hot-encoded row
Aadj_T: the transpose of the adjacency matrix
num_powers (int): the adjacency number of powers to compute
returns:
A matrix of the shape (num_powers, Aadj_T.shape[1]) of
the specified row of the first num_powers of the adjacency matrix.
"""
# make sure the transpose of the adjacency is used
# tensorflow requires that the sparse matrix is the first operand
partial_power = tf.reshape(
tf.sparse.to_dense(one_hot_encoded_row), shape=(1, Aadj_T.shape[1])
)
partial_powers_list = []
for i in range(num_powers):
partial_power = K.transpose(K.dot(Aadj_T, K.transpose(partial_power)))
partial_powers_list.append(partial_power)
return K.squeeze(tf.stack(partial_powers_list, axis=1), axis=0)
def _select_row_from_sparse_tensor(one_hot_encoded_row, sp_tensor_T):
"""
This function gathers the row specified in one_hot_encoded_row from the input sparse matrix
Args:
one_hot_encoded_row: one-hot-encoded row
sp_tensor_T: the transpose of the sparse matrix
returns:
The specified row from sp_tensor_T.
"""
one_hot_encoded_row = tf.reshape(
tf.sparse.to_dense(one_hot_encoded_row), shape=(1, sp_tensor_T.shape[1])
)
row_T = K.dot(sp_tensor_T, K.transpose(one_hot_encoded_row))
row = K.transpose(row_T)
return row