# Copyright (c) 2020 PaddlePaddle 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 math import paddle.fluid as fluid from paddlerec.core.utils import envs from paddlerec.core.model import ModelBase class Model(ModelBase): def __init__(self, config): ModelBase.__init__(self, config) def _init_hyper_parameters(self): self.sparse_feature_number = envs.get_global_env( "hyper_parameters.sparse_feature_number", None) self.sparse_feature_dim = envs.get_global_env( "hyper_parameters.sparse_feature_dim", None) self.num_field = envs.get_global_env("hyper_parameters.num_field", None) self.d_model = envs.get_global_env("hyper_parameters.d_model", None) self.d_key = envs.get_global_env("hyper_parameters.d_key", None) self.d_value = envs.get_global_env("hyper_parameters.d_value", None) self.n_head = envs.get_global_env("hyper_parameters.n_head", None) self.dropout_rate = envs.get_global_env( "hyper_parameters.dropout_rate", 0) self.n_interacting_layers = envs.get_global_env( "hyper_parameters.n_interacting_layers", 1) def multi_head_attention(self, queries, keys, values, d_key, d_value, d_model, n_head, dropout_rate): keys = queries if keys is None else keys values = keys if values is None else values if not (len(queries.shape) == len(keys.shape) == len(values.shape) == 3 ): raise ValueError( "Inputs: quries, keys and values should all be 3-D tensors.") def __compute_qkv(queries, keys, values, n_head, d_key, d_value): """ Add linear projection to queries, keys, and values. """ q = fluid.layers.fc(input=queries, size=d_key * n_head, bias_attr=False, num_flatten_dims=2) k = fluid.layers.fc(input=keys, size=d_key * n_head, bias_attr=False, num_flatten_dims=2) v = fluid.layers.fc(input=values, size=d_value * n_head, bias_attr=False, num_flatten_dims=2) return q, k, v def __split_heads_qkv(queries, keys, values, n_head, d_key, d_value): """ Reshape input tensors at the last dimension to split multi-heads and then transpose. Specifically, transform the input tensor with shape [bs, max_sequence_length, n_head * hidden_dim] to the output tensor with shape [bs, n_head, max_sequence_length, hidden_dim]. """ # The value 0 in shape attr means copying the corresponding dimension # size of the input as the output dimension size. reshaped_q = fluid.layers.reshape( x=queries, shape=[0, 0, n_head, d_key], inplace=True) # permuate the dimensions into: # [batch_size, n_head, max_sequence_len, hidden_size_per_head] q = fluid.layers.transpose(x=reshaped_q, perm=[0, 2, 1, 3]) # For encoder-decoder attention in inference, insert the ops and vars # into global block to use as cache among beam search. reshaped_k = fluid.layers.reshape( x=keys, shape=[0, 0, n_head, d_key], inplace=True) k = fluid.layers.transpose(x=reshaped_k, perm=[0, 2, 1, 3]) reshaped_v = fluid.layers.reshape( x=values, shape=[0, 0, n_head, d_value], inplace=True) v = fluid.layers.transpose(x=reshaped_v, perm=[0, 2, 1, 3]) return q, k, v def scaled_dot_product_attention(q, k, v, d_key, dropout_rate): """ Scaled Dot-Product Attention """ product = fluid.layers.matmul( x=q, y=k, transpose_y=True, alpha=d_key**-0.5) weights = fluid.layers.softmax(product) if dropout_rate: weights = fluid.layers.dropout( weights, dropout_prob=dropout_rate, seed=None, is_test=False) out = fluid.layers.matmul(weights, v) return out def __combine_heads(x): """ Transpose and then reshape the last two dimensions of inpunt tensor x so that it becomes one dimension, which is reverse to __split_heads. """ if len(x.shape) != 4: raise ValueError("Input(x) should be a 4-D Tensor.") trans_x = fluid.layers.transpose(x, perm=[0, 2, 1, 3]) # The value 0 in shape attr means copying the corresponding dimension # size of the input as the output dimension size. return fluid.layers.reshape( x=trans_x, shape=[0, 0, trans_x.shape[2] * trans_x.shape[3]], inplace=True) q, k, v = __compute_qkv(queries, keys, values, n_head, d_key, d_value) q, k, v = __split_heads_qkv(q, k, v, n_head, d_key, d_value) ctx_multiheads = scaled_dot_product_attention(q, k, v, self.d_model, dropout_rate) out = __combine_heads(ctx_multiheads) return out def interacting_layer(self, x): attention_out = self.multi_head_attention( x, None, None, self.d_key, self.d_value, self.d_model, self.n_head, self.dropout_rate) W_0_x = fluid.layers.fc(input=x, size=self.d_model, bias_attr=False, num_flatten_dims=2) res_out = fluid.layers.relu(attention_out + W_0_x) return res_out def net(self, inputs, is_infer=False): init_value_ = 0.1 is_distributed = True if envs.get_trainer() == "CtrTrainer" else False # ------------------------- network input -------------------------- raw_feat_idx = self._sparse_data_var[1] raw_feat_value = self._dense_data_var[0] self.label = self._sparse_data_var[0] feat_idx = raw_feat_idx feat_value = fluid.layers.reshape( raw_feat_value, [-1, self.num_field, 1]) # None * num_field * 1 # ------------------------- Embedding -------------------------- feat_embeddings_re = fluid.embedding( input=feat_idx, is_sparse=True, is_distributed=is_distributed, dtype='float32', size=[self.sparse_feature_number + 1, self.sparse_feature_dim], padding_idx=0, param_attr=fluid.ParamAttr( initializer=fluid.initializer.TruncatedNormalInitializer( loc=0.0, scale=init_value_ / math.sqrt(float(self.sparse_feature_dim))))) feat_embeddings = fluid.layers.reshape( feat_embeddings_re, shape=[-1, self.num_field, self.sparse_feature_dim ]) # None * num_field * embedding_size # None * num_field * embedding_size feat_embeddings = feat_embeddings * feat_value inter_input = feat_embeddings # ------------------------- interacting layer -------------------------- for _ in range(self.n_interacting_layers): interacting_layer_out = self.interacting_layer(inter_input) inter_input = interacting_layer_out # ------------------------- DNN -------------------------- dnn_input = fluid.layers.flatten(interacting_layer_out, axis=1) y_dnn = fluid.layers.fc( input=dnn_input, size=1, act=None, param_attr=fluid.ParamAttr( initializer=fluid.initializer.TruncatedNormalInitializer( loc=0.0, scale=init_value_)), bias_attr=fluid.ParamAttr( initializer=fluid.initializer.TruncatedNormalInitializer( loc=0.0, scale=init_value_))) self.predict = fluid.layers.sigmoid(y_dnn) cost = fluid.layers.log_loss( input=self.predict, label=fluid.layers.cast(self.label, "float32")) avg_cost = fluid.layers.reduce_sum(cost) self._cost = avg_cost predict_2d = fluid.layers.concat([1 - self.predict, self.predict], 1) label_int = fluid.layers.cast(self.label, 'int64') auc_var, batch_auc_var, _ = fluid.layers.auc(input=predict_2d, label=label_int, slide_steps=0) self._metrics["AUC"] = auc_var self._metrics["BATCH_AUC"] = batch_auc_var if is_infer: self._infer_results["AUC"] = auc_var