def inference(inputs, nb_classes, nb_nodes, training, attn_drop, ffd_drop,
bias_mat, hid_units, n_heads, activation=tf.nn.elu,
residual=False):
attns = []
for _ in range(n_heads[0]):
attns.append(layers.sp_attn_head(inputs,
adj_mat=bias_mat,
out_sz=hid_units[0], activation=activation, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=False))
h_1 = tf.concat(attns, axis=-1)
h_2 = h_1
for i in range(1, len(hid_units)):
h_old = h_1
attns = []
for _ in range(n_heads[i]):
attns.append(layers.sp_attn_head(h_1,
adj_mat=bias_mat,
out_sz=hid_units[i], activation=activation, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=residual))
h_1 = tf.concat(attns, axis=-1)
out = []
for i in range(n_heads[-1]):
out.append(layers.sp_attn_head(h_1, adj_mat=bias_mat,
out_sz=nb_classes, activation=lambda x: x, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=False))
logits = tf.add_n(out) / n_heads[-1]
return logits, h_2
def inference(inputs,
nb_classes,
nb_nodes,
training,
attn_drop,
ffd_drop,
bias_mat,
adj_all_mat,
adj_neig_mat,
N_target_mat,
hid_units,
n_heads,
activation=tf.nn.elu,
residual=False):
out = []
for i in range(n_heads[-1]):
out.append(
layers.sp_attn_head(inputs,
adj_mat=bias_mat,
adj_all_mat=adj_all_mat,
adj_neig_mat=adj_neig_mat,
N_target_mat=N_target_mat,
out_sz=nb_classes,
activation=lambda x: x,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=False))
logits = tf.add_n(out) / n_heads[-1]
return logits
def encoder(inputs_list,
nb_nodes,
training,
attn_drop,
ffd_drop,
bias_mat_local_list,
bias_mat_global_list,
hid_units,
n_heads,
mp_att_size=16,
activation=tf.nn.elu,
residual=False):
embed_list = []
for inputs, bias_mat_local, bias_mat_global in zip(
inputs_list, bias_mat_local_list, bias_mat_global_list):
attns = []
for _ in range(n_heads):
attn_temp = layers.sp_attn_head(inputs,
adj_mat_local=bias_mat_local,
adj_mat_global=bias_mat_global,
out_sz=hid_units[0],
activation=activation,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=False)
attns.append(attn_temp)
h_1 = tf.concat(attns, axis=-1)
embed_list.append(tf.expand_dims(tf.squeeze(h_1), axis=1))
multi_embed = tf.concat(embed_list, axis=1)
final_embed, alpha = layers.SimpleAttLayer(multi_embed,
mp_att_size,
time_major=False,
return_alphas=True)
return final_embed
def attn_h_1(inputs,
nb_classes,
nb_nodes,
training,
attn_drop,
ffd_drop,
bias_mat,
hid_units,
n_heads,
activation=tf.nn.elu,
residual=False):
attns = []
for _ in range(n_heads[0]):
attns.append(
layers.sp_attn_head(inputs,
adj_mat=bias_mat,
out_sz=hid_units[0],
activation=activation,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=False))
h_1 = tf.concat(attns, axis=-1)
for i in range(1, len(hid_units)):
h_old = h_1
attns = []
for _ in range(n_heads[i]):
attns.append(
layers.sp_attn_head(h_1,
adj_mat=bias_mat,
out_sz=hid_units[i],
activation=activation,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=residual))
h_1 = tf.concat(attns, axis=-1)
return h_1
def inference(inputs, nb_classes, nb_nodes, training, attn_drop, ffd_drop,
bias_mat, adj_hop1_all_mat, adj_hop2_all_mat, adj_hop1_neig_mat,adj_hop2_neig_mat, N_hop1_all_mat,N_hop2_all_mat,
hid_units, n_heads, activation=tf.nn.elu,
residual=False):
attns = []
for _ in range(n_heads[0]):
attns.append(layers.sp_attn_head(inputs, adj_mat=bias_mat,
adj_hop1_all_mat=adj_hop1_all_mat, adj_hop2_all_mat=adj_hop2_all_mat,
adj_hop1_neig_mat=adj_hop1_neig_mat,adj_hop2_neig_mat=adj_hop2_neig_mat,
N_hop1_all_mat=N_hop1_all_mat, N_hop2_all_mat=N_hop2_all_mat,
out_sz=hid_units[0], activation=activation, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=False))
h_1 = tf.concat(attns, axis=-1)
out = []
for i in range(n_heads[-1]):
out.append(layers.sp_attn_head(h_1, adj_mat=bias_mat,
adj_hop1_all_mat=adj_hop1_all_mat, adj_hop2_all_mat=adj_hop2_all_mat,
adj_hop1_neig_mat=adj_hop1_neig_mat,adj_hop2_neig_mat=adj_hop2_neig_mat,
N_hop1_all_mat=N_hop1_all_mat, N_hop2_all_mat=N_hop2_all_mat,
out_sz=nb_classes, activation=lambda x: x, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=False))
logits = tf.add_n(out) / n_heads[-1]
bi = []
bi.append(layers.Bilinear(inputs, adj_mat=bias_mat,
adj_hop1_all_mat=adj_hop1_all_mat, adj_hop2_all_mat=adj_hop2_all_mat,
adj_hop1_neig_mat=adj_hop1_neig_mat,adj_hop2_neig_mat=adj_hop2_neig_mat,
N_hop1_all_mat=N_hop1_all_mat, N_hop2_all_mat=N_hop2_all_mat,
out_sz=nb_classes, activation=lambda x: x, nb_nodes=nb_nodes,
in_drop=ffd_drop, coef_drop=attn_drop, residual=False))
bi_logits = tf.add_n(bi)
output = (1. - FLAGS.alpha) * logits + FLAGS.alpha * bi_logits
return output
def inference(inputs,
nb_classes,
nb_nodes,
training,
attn_drop,
ffd_drop,
nnz,
bias_mat,
hid_units,
n_heads,
activation=tf.nn.elu,
intra_drop=None,
intra_activation=None,
scheme_norm=None,
scheme_init_std=None,
residual=False,
use_bias=True):
attns = []
for _ in range(n_heads[0]):
attns.append(
layers.sp_attn_head(inputs,
adj_mat=bias_mat,
out_sz=hid_units[0],
activation=activation,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=False,
intra_drop=intra_drop,
intra_activation=intra_activation,
scheme_norm=scheme_norm,
scheme_init_std=scheme_init_std,
use_bias=use_bias))
h_1 = tf.concat(attns, axis=-1)
for i in range(1, len(hid_units)):
h_old = h_1
attns = []
for _ in range(n_heads[i]):
attns.append(
layers.sp_attn_head(h_1,
adj_mat=bias_mat,
out_sz=hid_units[i],
activation=activation,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=residual,
intra_drop=intra_drop,
intra_activation=intra_activation,
scheme_norm=scheme_norm,
scheme_init_std=scheme_init_std,
use_bias=use_bias))
h_1 = tf.concat(attns, axis=-1)
out = []
for i in range(n_heads[-1]):
out.append(
layers.sp_attn_head(h_1,
adj_mat=bias_mat,
out_sz=nb_classes,
activation=lambda x: x,
nb_nodes=nb_nodes,
in_drop=ffd_drop,
coef_drop=attn_drop,
residual=False,
intra_drop=intra_drop,
intra_activation=intra_activation,
scheme_norm=scheme_norm,
scheme_init_std=scheme_init_std,
use_bias=use_bias))
logits = tf.add_n(out) / n_heads[-1]
return logits
