def __init__(self):
super(Decoder, self).__init__()
self.attention_network = Attention()
# Decoder
self.embedding = nn.Embedding(
config.vocab_size,
config.emb_dim,
weight_attr=paddle.ParamAttr(initializer=I.Normal(
std=config.trunc_norm_init_std)))
self.x_context = nn.Linear(config.hidden_dim * 2 + config.emb_dim,
config.emb_dim)
self.lstm = nn.LSTM(
config.emb_dim,
config.hidden_dim,
num_layers=1,
direction='forward',
weight_ih_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-config.rand_unif_init_mag,
high=config.rand_unif_init_mag)),
bias_ih_attr=paddle.ParamAttr(initializer=I.Constant(value=0.0)))
if config.pointer_gen:
self.p_gen_linear = nn.Linear(
config.hidden_dim * 4 + config.emb_dim, 1)
self.out1 = nn.Linear(config.hidden_dim * 3, config.hidden_dim)
self.out2 = nn.Linear(
config.hidden_dim,
config.vocab_size,
weight_attr=paddle.ParamAttr(initializer=I.Normal(
std=config.trunc_norm_init_std)))
def __init__(self, with_lstm=True, n_group=1, in_channels=3):
super(ResNet_ASTER, self).__init__()
self.with_lstm = with_lstm
self.n_group = n_group
self.layer0 = nn.Sequential(
nn.Conv2D(in_channels,
32,
kernel_size=(3, 3),
stride=1,
padding=1,
bias_attr=False), nn.BatchNorm2D(32), nn.ReLU())
self.inplanes = 32
self.layer1 = self._make_layer(32, 3, [2, 2]) # [16, 50]
self.layer2 = self._make_layer(64, 4, [2, 2]) # [8, 25]
self.layer3 = self._make_layer(128, 6, [2, 1]) # [4, 25]
self.layer4 = self._make_layer(256, 6, [2, 1]) # [2, 25]
self.layer5 = self._make_layer(512, 3, [2, 1]) # [1, 25]
if with_lstm:
self.rnn = nn.LSTM(512, 256, direction="bidirect", num_layers=2)
self.out_channels = 2 * 256
else:
self.out_channels = 512
def __init__(self):
super(Encoder, self).__init__()
# Initialized embeddings
self.embedding = nn.Embedding(
config.vocab_size,
config.emb_dim,
weight_attr=paddle.ParamAttr(initializer=I.Normal(
std=config.trunc_norm_init_std)))
# Initialized lstm weights
self.lstm = nn.LSTM(
config.emb_dim,
config.hidden_dim,
num_layers=1,
direction='bidirect',
weight_ih_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-config.rand_unif_init_mag,
high=config.rand_unif_init_mag)),
bias_ih_attr=paddle.ParamAttr(initializer=I.Constant(value=0.0)))
# Initialized linear weights
self.W_h = nn.Linear(config.hidden_dim * 2,
config.hidden_dim * 2,
bias_attr=False)
def __init__(self, in_channels, hidden_size):
super(EncoderWithRNN, self).__init__()
self.out_channels = hidden_size * 2
self.lstm = nn.LSTM(in_channels,
hidden_size,
direction='bidirectional',
num_layers=2)
def __init__(self,
attention_layer,
vocab_size,
num_classes,
emb_dim=128,
lstm_hidden_size=196,
fc_hidden_size=96,
lstm_layers=1,
dropout_rate=0.0,
padding_idx=0):
super().__init__()
self.padding_idx = padding_idx
self.embedder = nn.Embedding(
num_embeddings=vocab_size,
embedding_dim=emb_dim,
padding_idx=padding_idx)
self.bilstm = nn.LSTM(
input_size=emb_dim,
hidden_size=lstm_hidden_size,
num_layers=lstm_layers,
dropout=dropout_rate,
direction='bidirect')
self.attention = attention_layer
if isinstance(attention_layer, SelfAttention):
self.fc = nn.Linear(lstm_hidden_size, fc_hidden_size)
elif isinstance(attention_layer, SelfInteractiveAttention):
self.fc = nn.Linear(lstm_hidden_size * 2, fc_hidden_size)
else:
raise RuntimeError("Unknown attention type %s." %
attention_layer.__class__.__name__)
self.output_layer = nn.Linear(fc_hidden_size, num_classes)
def __init__(self,
in_channels,
num_chars=92,
visual_dim=16,
fusion_dim=1024,
node_input=32,
node_embed=256,
edge_input=5,
edge_embed=256,
num_gnn=2,
num_classes=26,
bidirectional=False):
super().__init__()
self.fusion = Block([visual_dim, node_embed], node_embed, fusion_dim)
self.node_embed = nn.Embedding(num_chars, node_input, 0)
hidden = node_embed // 2 if bidirectional else node_embed
self.rnn = nn.LSTM(input_size=node_input,
hidden_size=hidden,
num_layers=1)
self.edge_embed = nn.Linear(edge_input, edge_embed)
self.gnn_layers = nn.LayerList(
[GNNLayer(node_embed, edge_embed) for _ in range(num_gnn)])
self.node_cls = nn.Linear(node_embed, num_classes)
self.edge_cls = nn.Linear(edge_embed, 2)
def __init__(self, n_mels, num_layers, hidden_size, output_size):
super().__init__()
self.lstm = nn.LSTM(n_mels, hidden_size, num_layers)
self.linear = nn.Linear(hidden_size, output_size)
self.similarity_weight = self.create_parameter(
[1], default_initializer=I.Constant(10.))
self.similarity_bias = self.create_parameter(
[1], default_initializer=I.Constant(-5.))
def __init__(self, vocab_size, embed_dim, hidden_size, num_layers):
super(Seq2SeqEncoder, self).__init__()
self.embedder = nn.Embedding(vocab_size, embed_dim)
self.lstm = nn.LSTM(input_size=embed_dim,
hidden_size=hidden_size,
num_layers=num_layers,
dropout=0.2 if num_layers > 1 else 0.)
def __init__(self,
hidden_size,
vocab_size,
class_num=2,
num_steps=128,
num_layers=1,
init_scale=0.1,
dropout=None):
# 参数含义如下：
# 1.hidden_size，表示embedding-size，hidden和cell向量的维度
# 2.vocab_size，模型可以考虑的词表大小
# 3.class_num，情感类型个数，可以是2分类，也可以是多分类
# 4.num_steps，表示这个情感分析模型最大可以考虑的句子长度
# 5.num_layers，表示网络的层数
# 6.init_scale，表示网络内部的参数的初始化范围
# 长短时记忆网络内部用了很多Tanh，Sigmoid等激活函数，这些函数对数值精度非常敏感，
# 因此我们一般只使用比较小的初始化范围，以保证效果
super(SentimentClassifier, self).__init__()
self.hidden_size = hidden_size
self.vocab_size = vocab_size
self.class_num = class_num
self.init_scale = init_scale
self.num_layers = num_layers
self.num_steps = num_steps
self.dropout = dropout
# 声明一个embedding层，用来把句子中的每个词转换为向量
self.embedding = nn.Embedding(
num_embeddings=vocab_size,
embedding_dim=hidden_size,
sparse=False,
weight_attr=paddle.ParamAttr(
initializer=paddle.nn.initializer.Uniform(low=-init_scale,
high=init_scale)))
# 声明一个LSTM模型，用来把每个句子抽象成向量
self.simple_lstm_rnn = nn.LSTM(input_size=hidden_size,
hidden_size=hidden_size,
num_layers=num_layers)
# 在得到一个句子的向量表示后，需要根据这个向量表示对这个句子进行分类
# 一般来说，可以把这个句子的向量表示乘以一个大小为[self.hidden_size, self.class_num]的W参数，
# 并加上一个大小为[self.class_num]的b参数，从而达到把句子向量映射到分类结果的目的
# 我们需要声明最终在使用句子向量映射到具体情感类别过程中所需要使用的参数
# 这个参数的大小一般是[self.hidden_size, self.class_num]
self.cls_fc = nn.Linear(in_features=self.hidden_size,
out_features=self.class_num,
weight_attr=None,
bias_attr=None)
self.dropout_layer = nn.Dropout(p=self.dropout,
mode='upscale_in_train')
def __init__(self, n_chars, n_embed, n_out, pad_index=0):
super(CharLSTMEncoder, self).__init__()
self.n_chars = n_chars
self.n_embed = n_embed
self.n_out = n_out
self.pad_index = pad_index
# the embedding layer
self.embed = nn.Embedding(num_embeddings=n_chars,
embedding_dim=n_embed)
# the lstm layer
self.lstm = nn.LSTM(input_size=n_embed,
hidden_size=n_out // 2,
direction="bidirectional")
def __init__(self, vocabulary, image_height, channel=1):
super(Model, self).__init__()
assert image_height % 32 == 0, 'image Height has to be a multiple of 32'
self.conv1 = nn.Conv2D(in_channels=channel, out_channels=64, kernel_size=3, stride=1, padding=1)
self.relu1 = nn.ReLU()
self.pool1 = nn.MaxPool2D(kernel_size=2, stride=2)
self.conv2 = nn.Conv2D(in_channels=64, out_channels=128, kernel_size=3, stride=1, padding=1)
self.relu2 = nn.ReLU()
self.pool2 = nn.MaxPool2D(kernel_size=2, stride=2)
self.conv3 = nn.Conv2D(in_channels=128, out_channels=256, kernel_size=3, stride=1, padding=1)
self.relu3 = nn.ReLU()
self.bn3 = nn.BatchNorm2D(256)
self.conv4 = nn.Conv2D(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1)
self.relu4 = nn.ReLU()
self.pool4 = nn.MaxPool2D(kernel_size=(2,2), stride=(2, 1), padding=(0, 1))
self.conv5 = nn.Conv2D(in_channels=256, out_channels=512, kernel_size=3, stride=1, padding=1)
self.relu5 = nn.ReLU()
self.bn5 = nn.BatchNorm2D(512)
self.conv6 = nn.Conv2D(in_channels=512, out_channels=512, kernel_size=3, stride=1, padding=1)
self.relu6 = nn.ReLU()
self.pool6 = nn.MaxPool2D(kernel_size=(2, 2), stride=(2, 1), padding=(0, 1))
self.conv7 = nn.Conv2D(in_channels=512, out_channels=512, kernel_size=2, stride=1, padding=0)
self.relu7 = nn.ReLU()
self.bn7 = nn.BatchNorm2D(512)
self.lstm1 = nn.LSTM(input_size=512, hidden_size=256, direction='bidirectional')
self.fc = nn.Linear(in_features=512, out_features=256)
self.lstm2 = nn.LSTM(input_size=256, hidden_size=256, direction='bidirectional')
self.output = nn.Linear(in_features=512, out_features=len(vocabulary))
def __init__(self,
vocab_size,
num_class,
emb_dim=512,
hidden_size=512,
n_lstm_layer=3,
is_bidirectory=True,
padding_idx=0,
epsilon=1e-5,
dropout_rate=0.1):
"""Init model
Args:
vocab_size (int): vocab size.
num_class (int): num of classes.
emb_dim (int, optional): embedding dimmension. Defaults to 512.
hidden_size (int, optional): hidden size. Defaults to 512.
n_lstm_layer (int, optional): number of lstm layer. Defaults to 3.
is_bidirectory (bool, optional): use bidirect lstm. Defaults to True.
padding_idx (int, optional): padding index. Defaults to 0.
epsilon (float, optional): epsilon. Defaults to 1e-5.
dropout_rate (float, optional): dropout rate. Defaults to 0.1.
"""
super().__init__()
self.padding_idx = padding_idx
self.embedder = nn.Embedding(vocab_size,
emb_dim,
padding_idx=padding_idx)
self.layer_norm = nn.LayerNorm(normalized_shape=emb_dim,
epsilon=epsilon)
self.dropout = nn.Dropout(p=dropout_rate)
direction = 'bidirectional' if is_bidirectory else 'forward'
self.lstm_encoder = nn.LSTM(emb_dim,
hidden_size,
num_layers=n_lstm_layer,
direction=direction)
# kernel_size = (5, hidden_size * 2) if is_bidirectory else (5, hidden_size)
in_channels = hidden_size * 2 if is_bidirectory else hidden_size
self.conv_encoder = nn.Conv1D(in_channels=in_channels,
out_channels=hidden_size,
kernel_size=5,
padding=2)
self.output_layer = nn.Conv1D(in_channels=hidden_size,
out_channels=num_class,
kernel_size=3,
padding=1)
def __init__(self,
n_words,
pad_index,
lstm_by_wp_embed_size=200,
n_embed=300,
n_lstm_hidden=300,
n_lstm_layers=3):
super(LSTMByWPEncoder, self).__init__()
self.pad_index = pad_index
self.word_embed = nn.Embedding(n_words, lstm_by_wp_embed_size)
self.lstm = nn.LSTM(input_size=lstm_by_wp_embed_size,
hidden_size=n_lstm_hidden,
num_layers=n_lstm_layers,
direction="bidirectional")
self.mlp_input_size = n_lstm_hidden * 2
def __init__(self,
batch_size,
input_size,
hidden_size,
num_layers,
dropout,
task='pre-train'):
super(ELMoBiLM, self).__init__()
self._num_layers = num_layers
self._dropout = dropout
self._task = task
self._lstm_layers = []
for direction in ['forward', 'backward']:
layers = []
for i in range(num_layers):
lstm = nn.LSTM(input_size=input_size,
hidden_size=hidden_size,
num_layers=1,
direction='forward',
weight_hh_attr=paddle.ParamAttr(
initializer=I.XavierUniform()),
weight_ih_attr=paddle.ParamAttr(
initializer=I.XavierUniform()),
bias_hh_attr=False,
bias_ih_attr=paddle.ParamAttr(
initializer=I.Constant(value=0.0)))
self.add_sublayer('{}_lstm_layer_{}'.format(direction, i),
lstm)
hidden_state = paddle.zeros(shape=[1, batch_size, hidden_size],
dtype='float32')
cell_state = paddle.zeros(shape=[1, batch_size, hidden_size],
dtype='float32')
layers.append({
'lstm': lstm,
'hidden_state': hidden_state,
'cell_state': cell_state
})
self._lstm_layers.append(layers)
if dropout:
self._dropout_layer = nn.Dropout(p=dropout)
def __init__(self,
vocab_size,
embed_dim,
hidden_size,
num_layers,
dropout_prob=0.,
init_scale=0.1):
super(Seq2SeqEncoder, self).__init__()
self.embedder = nn.Embedding(
vocab_size,
embed_dim,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.lstm = nn.LSTM(input_size=embed_dim,
hidden_size=hidden_size,
num_layers=num_layers,
direction="forward",
dropout=dropout_prob if num_layers > 1 else 0.)
def __init__(self,
embed_dim,
hidden_size,
vocab_size,
output_dim,
vocab_path,
padding_idx=0,
num_layers=1,
dropout_prob=0.0,
init_scale=0.1,
embedding_name=None):
super(BiLSTM, self).__init__()
if embedding_name is not None:
self.embedder = TokenEmbedding(embedding_name,
extended_vocab_path=vocab_path,
keep_extended_vocab_only=True)
embed_dim = self.embedder.embedding_dim
else:
self.embedder = nn.Embedding(vocab_size, embed_dim, padding_idx)
self.lstm = nn.LSTM(embed_dim,
hidden_size,
num_layers,
'bidirectional',
dropout=dropout_prob)
self.fc = nn.Linear(
hidden_size * 2,
hidden_size,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.fc_1 = nn.Linear(
hidden_size * 8,
hidden_size,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.output_layer = nn.Linear(
hidden_size,
output_dim,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
def __init__(self,
input_size,
hidden_size,
num_layers=1,
direction="forward",
dropout=0.0,
pooling_type=None,
**kwargs):
super().__init__()
self._input_size = input_size
self._hidden_size = hidden_size
self._direction = direction
self._pooling_type = pooling_type
self.lstm_layer = nn.LSTM(input_size=input_size,
hidden_size=hidden_size,
num_layers=num_layers,
direction=direction,
dropout=dropout,
**kwargs)
def __init__(self,
feat,
n_feats,
n_words,
pad_index=0,
feat_pad_index=0,
n_char_embed=50,
n_feat_embed=60,
n_lstm_char_embed=100,
n_embed=300,
embed_dropout=0.33,
n_lstm_hidden=300,
n_lstm_layers=3,
lstm_dropout=0.33):
super(LSTMEncoder, self).__init__()
self.pad_index = pad_index
if feat == "char":
self.feat_embed = CharLSTMEncoder(
n_chars=n_feats,
n_embed=n_char_embed,
n_out=n_lstm_char_embed,
pad_index=feat_pad_index,
)
feat_embed_size = n_lstm_char_embed
else:
self.feat_embed = nn.Embedding(num_embeddings=n_feats,
embedding_dim=n_feat_embed)
feat_embed_size = n_feat_embed
self.word_embed = nn.Embedding(num_embeddings=n_words,
embedding_dim=n_embed)
self.embed_dropout = IndependentDropout(p=embed_dropout)
self.lstm = nn.LSTM(input_size=n_embed + feat_embed_size,
hidden_size=n_lstm_hidden,
num_layers=n_lstm_layers,
dropout=lstm_dropout,
direction="bidirectional")
self.lstm_dropout = SharedDropout(p=lstm_dropout)
self.mlp_input_size = n_lstm_hidden * 2
def __init__(self,
vocab_size,
emb_dim=128,
hidden_size=1024,
n_layers=3,
padding_idx=0,
epsilon=1e-5,
dropout_rate=0.1):
"""
__init__
"""
super(LstmEncoderModel, self).__init__()
self.padding_idx = padding_idx
self.embedding = nn.Embedding(vocab_size,
emb_dim,
padding_idx=padding_idx)
self.dropout = nn.Dropout(p=dropout_rate)
self.lstm_encoder = nn.LSTM(emb_dim,
hidden_size,
num_layers=n_layers,
direction="bidirectional")
def __init__(self,
vocab_size,
embed_dim,
hidden_size,
num_layers,
init_scale=0.1,
enc_dropout=0.):
super(LSTMEncoder, self).__init__()
self.src_embedder = nn.Embedding(
vocab_size,
embed_dim,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.lstm = nn.LSTM(input_size=embed_dim,
hidden_size=hidden_size,
num_layers=num_layers,
dropout=enc_dropout)
if enc_dropout > 0.0:
self.dropout = nn.Dropout(enc_dropout)
else:
self.dropout = None
def __init__(self,
enc_bi_rnn=False,
enc_drop_rnn=0.1,
enc_gru=False,
d_model=512,
d_enc=512,
mask=True,
**kwargs):
super().__init__()
assert isinstance(enc_bi_rnn, bool)
assert isinstance(enc_drop_rnn, (int, float))
assert 0 <= enc_drop_rnn < 1.0
assert isinstance(enc_gru, bool)
assert isinstance(d_model, int)
assert isinstance(d_enc, int)
assert isinstance(mask, bool)
self.enc_bi_rnn = enc_bi_rnn
self.enc_drop_rnn = enc_drop_rnn
self.mask = mask
# LSTM Encoder
if enc_bi_rnn:
direction = 'bidirectional'
else:
direction = 'forward'
kwargs = dict(input_size=d_model,
hidden_size=d_enc,
num_layers=2,
time_major=False,
dropout=enc_drop_rnn,
direction=direction)
if enc_gru:
self.rnn_encoder = nn.GRU(**kwargs)
else:
self.rnn_encoder = nn.LSTM(**kwargs)
# global feature transformation
encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
self.linear = nn.Linear(encoder_rnn_out_size, encoder_rnn_out_size)
def __init__(self,
vocab_size,
hidden_size,
batch_size,
num_layers=1,
init_scale=0.1,
dropout=0.0):
super(RnnLm, self).__init__()
self.hidden_size = hidden_size
self.num_layers = num_layers
self.init_scale = init_scale
self.batch_size = batch_size
self.reset_states()
self.embedder = nn.Embedding(
vocab_size,
hidden_size,
weight_attr=paddle.ParamAttr(initializer=I.Uniform(
low=-init_scale, high=init_scale)))
self.lstm = nn.LSTM(
input_size=hidden_size,
hidden_size=hidden_size,
num_layers=num_layers,
dropout=dropout,
weight_ih_attr=paddle.ParamAttr(initializer=I.Uniform(
low=-init_scale, high=init_scale)),
weight_hh_attr=paddle.ParamAttr(initializer=I.Uniform(
low=-init_scale, high=init_scale)))
self.fc = nn.Linear(
hidden_size,
vocab_size,
weight_attr=paddle.ParamAttr(initializer=I.Uniform(
low=-init_scale, high=init_scale)),
bias_attr=paddle.ParamAttr(initializer=I.Uniform(
low=-init_scale, high=init_scale)))
self.dropout = nn.Dropout(p=dropout)
def __init__(self, d_hidden: int, conv_layers: int, kernel_size: int,
p_dropout: float):
super().__init__()
k = math.sqrt(1.0 / (d_hidden * kernel_size))
self.conv_batchnorms = paddle.nn.LayerList([
Conv1dBatchNorm(
d_hidden,
d_hidden,
kernel_size,
stride=1,
padding=int((kernel_size - 1) / 2),
bias_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(low=-k, high=k)),
data_format='NLC') for i in range(conv_layers)
])
self.p_dropout = p_dropout
self.hidden_size = int(d_hidden / 2)
self.lstm = nn.LSTM(d_hidden,
self.hidden_size,
direction="bidirectional")
def __init__(self,
embed_dim,
hidden_size,
vocab_size,
output_dim,
padding_idx=0,
num_layers=1,
dropout_prob=0.0,
init_scale=0.1,
embed_weight=None):
super(BiLSTM, self).__init__()
self.embedder = nn.Embedding(vocab_size, embed_dim, padding_idx)
self.embedder.weight.set_value(
embed_weight) if embed_weight is not None else None
self.lstm = nn.LSTM(embed_dim,
hidden_size,
num_layers,
'bidirectional',
dropout=dropout_prob)
self.fc = nn.Linear(
hidden_size * 2,
hidden_size,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.fc_1 = nn.Linear(
hidden_size * 8,
hidden_size,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
self.output_layer = nn.Linear(
hidden_size,
output_dim,
weight_attr=paddle.ParamAttr(
initializer=I.Uniform(low=-init_scale, high=init_scale)))
def __init__(self,
vocab_size,
num_classes,
emb_dim=128,
padding_idx=0,
lstm_hidden_size=198,
direction='forward',
lstm_layers=1,
dropout_rate=0.0,
pooling_type=None,
fc_hidden_size=96):
super().__init__()
self.direction = direction
self.embedder = nn.Embedding(num_embeddings=vocab_size,
embedding_dim=emb_dim,
padding_idx=padding_idx)
# self.lstm_encoder = nlp.seq2vec.LSTMEncoder(emb_dim,
# lstm_hidden_size,
# num_layers=lstm_layers,
# direction=direction,
# dropout=dropout_rate,
# pooling_type=pooling_type)
self.lstm_layer = nn.LSTM(input_size=emb_dim,
hidden_size=lstm_hidden_size,
num_layers=lstm_layers,
direction=direction,
dropout=dropout_rate)
self.fc = nn.Linear(
lstm_hidden_size * (2 if direction == 'bidirect' else 1),
fc_hidden_size)
self.output_layer = nn.Linear(fc_hidden_size, num_classes)
self.softmax = nn.Softmax(axis=1)
def paddle_lstm():
np.random.seed(SEED)
x = np.random.rand(1, 80, 512).astype(np.float32)
# np.save('org.npy', x)
with fluid.dygraph.guard():
lstm = nn.LSTM(512, 256, num_layers=2, direction='bidirectional')
# sd = np.load('lstm.npy', allow_pickle=True).tolist()
# lstm.set_state_dict(sd)
state_dict = lstm.state_dict()
sd = OrderedDict()
for key, value in state_dict.items():
v = value.numpy()
print(key, value.shape, np.sum(v), np.mean(v), np.max(v),
np.min(v))
sd[key] = v
np.save('lstm.npy', sd)
inp = fluid.dygraph.to_variable(x)
ret, _ = lstm(inp)
print(len(ret))
return ret.numpy()
def __init__(
self,
user_size,
adgroup_size,
pid_size,
cms_segid_size,
cms_group_size,
final_gender_size,
age_level_size,
pvalue_level_size,
shopping_level_size,
occupation_size,
new_user_class_level_size,
campaign_size,
customer_size,
cate_size,
brand_size, # above is all sparse feat size
sparse_embed_size=4,
att_embedding_size=8,
sess_count=5,
sess_max_length=10,
l2_reg_embedding=1e-6):
super().__init__()
# feature size
self.user_size = user_size
self.adgroup_size = adgroup_size
self.pid_size = pid_size
self.cms_segid_size = cms_segid_size
self.cms_group_size = cms_group_size
self.final_gender_size = final_gender_size
self.age_level_size = age_level_size
self.pvalue_level_size = pvalue_level_size
self.shopping_level_size = shopping_level_size
self.occupation_size = occupation_size
self.new_user_class_level_size = new_user_class_level_size
self.campaign_size = campaign_size
self.customer_size = customer_size
self.cate_size = cate_size
self.brand_size = brand_size
# sparse embed size
self.sparse_embed_size = sparse_embed_size
# transform attention embed size
self.att_embedding_size = att_embedding_size
# hyper_parameters
self.sess_count = 5
self.sess_max_length = 10
# sparse embedding layer
self.userid_embeddings_var = paddle.nn.Embedding(
self.user_size,
self.sparse_embed_size,
sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.adgroup_embeddings_var = paddle.nn.Embedding(
self.adgroup_size,
self.sparse_embed_size,
sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.pid_embeddings_var = paddle.nn.Embedding(
self.pid_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.cmsid_embeddings_var = paddle.nn.Embedding(
self.cms_segid_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.cmsgroup_embeddings_var = paddle.nn.Embedding(
self.cms_group_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.gender_embeddings_var = paddle.nn.Embedding(
self.final_gender_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.age_embeddings_var = paddle.nn.Embedding(
self.age_level_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.pvalue_embeddings_var = paddle.nn.Embedding(
self.pvalue_level_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.shopping_embeddings_var = paddle.nn.Embedding(
self.shopping_level_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.occupation_embeddings_var = paddle.nn.Embedding(
self.occupation_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.new_user_class_level_embeddings_var = paddle.nn.Embedding(
self.new_user_class_level_size,
self.sparse_embed_size,
#sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.campaign_embeddings_var = paddle.nn.Embedding(
self.campaign_size,
self.sparse_embed_size,
sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.customer_embeddings_var = paddle.nn.Embedding(
self.customer_size,
self.sparse_embed_size,
sparse=True,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.cate_embeddings_var = paddle.nn.Embedding(
self.cate_size,
self.sparse_embed_size,
sparse=True,
padding_idx=0,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
self.brand_embeddings_var = paddle.nn.Embedding(
self.brand_size,
self.sparse_embed_size,
sparse=True,
padding_idx=0,
weight_attr=paddle.ParamAttr(
regularizer=paddle.regularizer.L2Decay(l2_reg_embedding),
initializer=nn.initializer.Normal(mean=0.0, std=0.0001)))
# sess interest extractor layer
self.position_encoding = PositionalEncoder(2 * self.sparse_embed_size)
self.transform = nn.TransformerEncoderLayer(
d_model=self.att_embedding_size,
nhead=8,
dim_feedforward=64,
weight_attr=self._get_weight_attr(),
bias_attr=False,
dropout=0.0)
# sess interest interacting layer
self.bilstm = nn.LSTM(2 * self.sparse_embed_size,
2 * self.sparse_embed_size,
num_layers=2,
direction='bidirectional')
# sess interest activating layer
self.transform_actpool = AttentionSequencePoolingLayer(
weight_normalization=True, name='transform')
self.lstm_actpool = AttentionSequencePoolingLayer(
weight_normalization=True, name='lstm')
# MLP moudle
self.mlp = MLP(mlp_hidden_units=[77, 200, 80])
def __init__(
self,
out_channels, # 90 + unknown + start + padding
enc_bi_rnn=False,
dec_bi_rnn=False,
dec_drop_rnn=0.0,
dec_gru=False,
d_model=512,
d_enc=512,
d_k=64,
pred_dropout=0.1,
max_text_length=30,
mask=True,
pred_concat=True,
**kwargs):
super().__init__()
self.num_classes = out_channels
self.enc_bi_rnn = enc_bi_rnn
self.d_k = d_k
self.start_idx = out_channels - 2
self.padding_idx = out_channels - 1
self.max_seq_len = max_text_length
self.mask = mask
self.pred_concat = pred_concat
encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
decoder_rnn_out_size = encoder_rnn_out_size * (int(dec_bi_rnn) + 1)
# 2D attention layer
self.conv1x1_1 = nn.Linear(decoder_rnn_out_size, d_k)
self.conv3x3_1 = nn.Conv2D(d_model,
d_k,
kernel_size=3,
stride=1,
padding=1)
self.conv1x1_2 = nn.Linear(d_k, 1)
# Decoder RNN layer
if dec_bi_rnn:
direction = 'bidirectional'
else:
direction = 'forward'
kwargs = dict(input_size=encoder_rnn_out_size,
hidden_size=encoder_rnn_out_size,
num_layers=2,
time_major=False,
dropout=dec_drop_rnn,
direction=direction)
if dec_gru:
self.rnn_decoder = nn.GRU(**kwargs)
else:
self.rnn_decoder = nn.LSTM(**kwargs)
# Decoder input embedding
self.embedding = nn.Embedding(self.num_classes,
encoder_rnn_out_size,
padding_idx=self.padding_idx)
# Prediction layer
self.pred_dropout = nn.Dropout(pred_dropout)
pred_num_classes = self.num_classes - 1
if pred_concat:
fc_in_channel = decoder_rnn_out_size + d_model + d_enc
else:
fc_in_channel = d_model
self.prediction = nn.Linear(fc_in_channel, pred_num_classes)
def func_test_layer_str(self):
module = nn.ELU(0.2)
self.assertEqual(str(module), 'ELU(alpha=0.2)')
module = nn.CELU(0.2)
self.assertEqual(str(module), 'CELU(alpha=0.2)')
module = nn.GELU(True)
self.assertEqual(str(module), 'GELU(approximate=True)')
module = nn.Hardshrink()
self.assertEqual(str(module), 'Hardshrink(threshold=0.5)')
module = nn.Hardswish(name="Hardswish")
self.assertEqual(str(module), 'Hardswish(name=Hardswish)')
module = nn.Tanh(name="Tanh")
self.assertEqual(str(module), 'Tanh(name=Tanh)')
module = nn.Hardtanh(name="Hardtanh")
self.assertEqual(str(module),
'Hardtanh(min=-1.0, max=1.0, name=Hardtanh)')
module = nn.PReLU(1, 0.25, name="PReLU", data_format="NCHW")
self.assertEqual(
str(module),
'PReLU(num_parameters=1, data_format=NCHW, init=0.25, dtype=float32, name=PReLU)'
)
module = nn.ReLU()
self.assertEqual(str(module), 'ReLU()')
module = nn.ReLU6()
self.assertEqual(str(module), 'ReLU6()')
module = nn.SELU()
self.assertEqual(
str(module),
'SELU(scale=1.0507009873554805, alpha=1.6732632423543772)')
module = nn.LeakyReLU()
self.assertEqual(str(module), 'LeakyReLU(negative_slope=0.01)')
module = nn.Sigmoid()
self.assertEqual(str(module), 'Sigmoid()')
module = nn.Hardsigmoid()
self.assertEqual(str(module), 'Hardsigmoid()')
module = nn.Softplus()
self.assertEqual(str(module), 'Softplus(beta=1, threshold=20)')
module = nn.Softshrink()
self.assertEqual(str(module), 'Softshrink(threshold=0.5)')
module = nn.Softsign()
self.assertEqual(str(module), 'Softsign()')
module = nn.Swish()
self.assertEqual(str(module), 'Swish()')
module = nn.Tanhshrink()
self.assertEqual(str(module), 'Tanhshrink()')
module = nn.ThresholdedReLU()
self.assertEqual(str(module), 'ThresholdedReLU(threshold=1.0)')
module = nn.LogSigmoid()
self.assertEqual(str(module), 'LogSigmoid()')
module = nn.Softmax()
self.assertEqual(str(module), 'Softmax(axis=-1)')
module = nn.LogSoftmax()
self.assertEqual(str(module), 'LogSoftmax(axis=-1)')
module = nn.Maxout(groups=2)
self.assertEqual(str(module), 'Maxout(groups=2, axis=1)')
module = nn.Linear(2, 4, name='linear')
self.assertEqual(
str(module),
'Linear(in_features=2, out_features=4, dtype=float32, name=linear)'
)
module = nn.Upsample(size=[12, 12])
self.assertEqual(
str(module),
'Upsample(size=[12, 12], mode=nearest, align_corners=False, align_mode=0, data_format=NCHW)'
)
module = nn.UpsamplingNearest2D(size=[12, 12])
self.assertEqual(
str(module),
'UpsamplingNearest2D(size=[12, 12], data_format=NCHW)')
module = nn.UpsamplingBilinear2D(size=[12, 12])
self.assertEqual(
str(module),
'UpsamplingBilinear2D(size=[12, 12], data_format=NCHW)')
module = nn.Bilinear(in1_features=5, in2_features=4, out_features=1000)
self.assertEqual(
str(module),
'Bilinear(in1_features=5, in2_features=4, out_features=1000, dtype=float32)'
)
module = nn.Dropout(p=0.5)
self.assertEqual(str(module),
'Dropout(p=0.5, axis=None, mode=upscale_in_train)')
module = nn.Dropout2D(p=0.5)
self.assertEqual(str(module), 'Dropout2D(p=0.5, data_format=NCHW)')
module = nn.Dropout3D(p=0.5)
self.assertEqual(str(module), 'Dropout3D(p=0.5, data_format=NCDHW)')
module = nn.AlphaDropout(p=0.5)
self.assertEqual(str(module), 'AlphaDropout(p=0.5)')
module = nn.Pad1D(padding=[1, 2], mode='constant')
self.assertEqual(
str(module),
'Pad1D(padding=[1, 2], mode=constant, value=0.0, data_format=NCL)')
module = nn.Pad2D(padding=[1, 0, 1, 2], mode='constant')
self.assertEqual(
str(module),
'Pad2D(padding=[1, 0, 1, 2], mode=constant, value=0.0, data_format=NCHW)'
)
module = nn.ZeroPad2D(padding=[1, 0, 1, 2])
self.assertEqual(str(module),
'ZeroPad2D(padding=[1, 0, 1, 2], data_format=NCHW)')
module = nn.Pad3D(padding=[1, 0, 1, 2, 0, 0], mode='constant')
self.assertEqual(
str(module),
'Pad3D(padding=[1, 0, 1, 2, 0, 0], mode=constant, value=0.0, data_format=NCDHW)'
)
module = nn.CosineSimilarity(axis=0)
self.assertEqual(str(module), 'CosineSimilarity(axis=0, eps=1e-08)')
module = nn.Embedding(10, 3, sparse=True)
self.assertEqual(str(module), 'Embedding(10, 3, sparse=True)')
module = nn.Conv1D(3, 2, 3)
self.assertEqual(str(module),
'Conv1D(3, 2, kernel_size=[3], data_format=NCL)')
module = nn.Conv1DTranspose(2, 1, 2)
self.assertEqual(
str(module),
'Conv1DTranspose(2, 1, kernel_size=[2], data_format=NCL)')
module = nn.Conv2D(4, 6, (3, 3))
self.assertEqual(str(module),
'Conv2D(4, 6, kernel_size=[3, 3], data_format=NCHW)')
module = nn.Conv2DTranspose(4, 6, (3, 3))
self.assertEqual(
str(module),
'Conv2DTranspose(4, 6, kernel_size=[3, 3], data_format=NCHW)')
module = nn.Conv3D(4, 6, (3, 3, 3))
self.assertEqual(
str(module),
'Conv3D(4, 6, kernel_size=[3, 3, 3], data_format=NCDHW)')
module = nn.Conv3DTranspose(4, 6, (3, 3, 3))
self.assertEqual(
str(module),
'Conv3DTranspose(4, 6, kernel_size=[3, 3, 3], data_format=NCDHW)')
module = nn.PairwiseDistance()
self.assertEqual(str(module), 'PairwiseDistance(p=2.0)')
module = nn.InstanceNorm1D(2)
self.assertEqual(str(module),
'InstanceNorm1D(num_features=2, epsilon=1e-05)')
module = nn.InstanceNorm2D(2)
self.assertEqual(str(module),
'InstanceNorm2D(num_features=2, epsilon=1e-05)')
module = nn.InstanceNorm3D(2)
self.assertEqual(str(module),
'InstanceNorm3D(num_features=2, epsilon=1e-05)')
module = nn.GroupNorm(num_channels=6, num_groups=6)
self.assertEqual(
str(module),
'GroupNorm(num_groups=6, num_channels=6, epsilon=1e-05)')
module = nn.LayerNorm([2, 2, 3])
self.assertEqual(
str(module),
'LayerNorm(normalized_shape=[2, 2, 3], epsilon=1e-05)')
module = nn.BatchNorm1D(1)
self.assertEqual(
str(module),
'BatchNorm1D(num_features=1, momentum=0.9, epsilon=1e-05, data_format=NCL)'
)
module = nn.BatchNorm2D(1)
self.assertEqual(
str(module),
'BatchNorm2D(num_features=1, momentum=0.9, epsilon=1e-05)')
module = nn.BatchNorm3D(1)
self.assertEqual(
str(module),
'BatchNorm3D(num_features=1, momentum=0.9, epsilon=1e-05, data_format=NCDHW)'
)
module = nn.SyncBatchNorm(2)
self.assertEqual(
str(module),
'SyncBatchNorm(num_features=2, momentum=0.9, epsilon=1e-05)')
module = nn.LocalResponseNorm(size=5)
self.assertEqual(
str(module),
'LocalResponseNorm(size=5, alpha=0.0001, beta=0.75, k=1.0)')
module = nn.AvgPool1D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'AvgPool1D(kernel_size=2, stride=2, padding=0)')
module = nn.AvgPool2D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'AvgPool2D(kernel_size=2, stride=2, padding=0)')
module = nn.AvgPool3D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'AvgPool3D(kernel_size=2, stride=2, padding=0)')
module = nn.MaxPool1D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'MaxPool1D(kernel_size=2, stride=2, padding=0)')
module = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'MaxPool2D(kernel_size=2, stride=2, padding=0)')
module = nn.MaxPool3D(kernel_size=2, stride=2, padding=0)
self.assertEqual(str(module),
'MaxPool3D(kernel_size=2, stride=2, padding=0)')
module = nn.AdaptiveAvgPool1D(output_size=16)
self.assertEqual(str(module), 'AdaptiveAvgPool1D(output_size=16)')
module = nn.AdaptiveAvgPool2D(output_size=3)
self.assertEqual(str(module), 'AdaptiveAvgPool2D(output_size=3)')
module = nn.AdaptiveAvgPool3D(output_size=3)
self.assertEqual(str(module), 'AdaptiveAvgPool3D(output_size=3)')
module = nn.AdaptiveMaxPool1D(output_size=16, return_mask=True)
self.assertEqual(
str(module), 'AdaptiveMaxPool1D(output_size=16, return_mask=True)')
module = nn.AdaptiveMaxPool2D(output_size=3, return_mask=True)
self.assertEqual(str(module),
'AdaptiveMaxPool2D(output_size=3, return_mask=True)')
module = nn.AdaptiveMaxPool3D(output_size=3, return_mask=True)
self.assertEqual(str(module),
'AdaptiveMaxPool3D(output_size=3, return_mask=True)')
module = nn.SimpleRNNCell(16, 32)
self.assertEqual(str(module), 'SimpleRNNCell(16, 32)')
module = nn.LSTMCell(16, 32)
self.assertEqual(str(module), 'LSTMCell(16, 32)')
module = nn.GRUCell(16, 32)
self.assertEqual(str(module), 'GRUCell(16, 32)')
module = nn.PixelShuffle(3)
self.assertEqual(str(module), 'PixelShuffle(upscale_factor=3)')
module = nn.SimpleRNN(16, 32, 2)
self.assertEqual(
str(module),
'SimpleRNN(16, 32, num_layers=2\n (0): RNN(\n (cell): SimpleRNNCell(16, 32)\n )\n (1): RNN(\n (cell): SimpleRNNCell(32, 32)\n )\n)'
)
module = nn.LSTM(16, 32, 2)
self.assertEqual(
str(module),
'LSTM(16, 32, num_layers=2\n (0): RNN(\n (cell): LSTMCell(16, 32)\n )\n (1): RNN(\n (cell): LSTMCell(32, 32)\n )\n)'
)
module = nn.GRU(16, 32, 2)
self.assertEqual(
str(module),
'GRU(16, 32, num_layers=2\n (0): RNN(\n (cell): GRUCell(16, 32)\n )\n (1): RNN(\n (cell): GRUCell(32, 32)\n )\n)'
)
module1 = nn.Sequential(
('conv1', nn.Conv2D(1, 20, 5)), ('relu1', nn.ReLU()),
('conv2', nn.Conv2D(20, 64, 5)), ('relu2', nn.ReLU()))
self.assertEqual(
str(module1),
'Sequential(\n '\
'(conv1): Conv2D(1, 20, kernel_size=[5, 5], data_format=NCHW)\n '\
'(relu1): ReLU()\n '\
'(conv2): Conv2D(20, 64, kernel_size=[5, 5], data_format=NCHW)\n '\
'(relu2): ReLU()\n)'
)
module2 = nn.Sequential(
nn.Conv3DTranspose(4, 6, (3, 3, 3)),
nn.AvgPool3D(kernel_size=2, stride=2, padding=0),
nn.Tanh(name="Tanh"), module1, nn.Conv3D(4, 6, (3, 3, 3)),
nn.MaxPool3D(kernel_size=2, stride=2, padding=0), nn.GELU(True))
self.assertEqual(
str(module2),
'Sequential(\n '\
'(0): Conv3DTranspose(4, 6, kernel_size=[3, 3, 3], data_format=NCDHW)\n '\
'(1): AvgPool3D(kernel_size=2, stride=2, padding=0)\n '\
'(2): Tanh(name=Tanh)\n '\
'(3): Sequential(\n (conv1): Conv2D(1, 20, kernel_size=[5, 5], data_format=NCHW)\n (relu1): ReLU()\n'\
' (conv2): Conv2D(20, 64, kernel_size=[5, 5], data_format=NCHW)\n (relu2): ReLU()\n )\n '\
'(4): Conv3D(4, 6, kernel_size=[3, 3, 3], data_format=NCDHW)\n '\
'(5): MaxPool3D(kernel_size=2, stride=2, padding=0)\n '\
'(6): GELU(approximate=True)\n)'
)
def __init__(self, in_channels, hidden_size):
super(LSTMLayer, self).__init__()
self.cell = nn.LSTM(in_channels,
hidden_size,
direction='bidirectional',
num_layers=2)
