基于CNN、LSTM、Transformer以及BERT各种神经网络进行情感分类项目源码.zip

# -*- coding: utf-8 -*- """ ------------------------------------------------- File Name： models Description : Author : Stephen.Lau date： 2019/3/25 ------------------------------------------------- Change Activity: 2019/3/25: ------------------------------------------------- """ # encoding: utf-8 """ @author: Alex @contact: ialexwwang@gmail.com @version: 0.1 @license: Apache Licence @file: deep_models.py @time: 2019-02-21 17:54 @Reference: https://github.com/zake7749/DeepToxic/blob/master/sotoxic/models/keras/model_zoo.py """ # from __future__ import absolute_import, division import keras from keras.layers import Bidirectional, Conv1D from keras.layers import Dense, Lambda, Flatten from keras.layers import Dropout, SpatialDropout1D from keras.layers import GlobalAveragePooling1D, GlobalMaxPooling1D, MaxPooling1D from keras.layers import concatenate from keras.models import Model from kashgari.layers import AttentionWeightedAverage, KMaxPooling, LSTMLayer, GRULayer from kashgari.tasks.classification.base_model import ClassificationModel class CNNModel(ClassificationModel): __architect_name__ = 'CNNModel' __base_hyper_parameters__ = { 'conv1d_layer': { 'filters': 128, 'kernel_size': 5, 'activation': 'relu' }, 'max_pool_layer': {}, 'dense_1_layer': { 'units': 64, 'activation': 'relu' }, 'activation_layer': { 'activation': 'softmax' }, 'optimizer': { 'module': 'keras.optimizers', 'name': 'Adam', 'params': { 'lr': 1e-3, 'decay': 0.0 } }, 'compile_params': { 'loss': 'categorical_crossentropy', # 'optimizer': 'adam', 'metrics': ['accuracy'] } } def build_model(self): base_model = self.embedding.model conv1d_layer = Conv1D(**self.hyper_parameters['conv1d_layer'])(base_model.output) max_pool_layer = GlobalMaxPooling1D(**self.hyper_parameters['max_pool_layer'])(conv1d_layer) dense_1_layer = Dense(**self.hyper_parameters['dense_1_layer'])(max_pool_layer) dense_2_layer = Dense(len(self.label2idx), **self.hyper_parameters['activation_layer'])(dense_1_layer) model = Model(base_model.inputs, dense_2_layer) optimizer = getattr(eval(self.hyper_parameters['optimizer']['module']), self.hyper_parameters['optimizer']['name'])( **self.hyper_parameters['optimizer']['params']) model.compile(optimizer=optimizer, **self.hyper_parameters['compile_params']) self.model = model self.model.summary() class BLSTMModel(ClassificationModel): __architect_name__ = 'BLSTMModel' __base_hyper_parameters__ = { 'lstm_layer': { 'units': 256, 'return_sequences': False }, 'activation_layer': { 'activation': 'softmax' }, 'optimizer': { 'module': 'keras.optimizers', 'name': 'Adam', 'params': { 'lr': 1e-3, 'decay': 0.0 } }, 'compile_params': { 'loss': 'categorical_crossentropy', # 'optimizer': 'adam', 'metrics': ['accuracy'] } } def build_model(self): base_model = self.embedding.model blstm_layer = Bidirectional(LSTMLayer(**self.hyper_parameters['lstm_layer']))(base_model.output) dense_layer = Dense(len(self.label2idx), **self.hyper_parameters['activation_layer'])(blstm_layer) output_layers = [dense_layer] model = Model(base_model.inputs, output_layers) optimizer = getattr(eval(self.hyper_parameters['optimizer']['module']), self.hyper_parameters['optimizer']['name'])( **self.hyper_parameters['optimizer']['params']) model.compile(optimizer=optimizer, **self.hyper_parameters['compile_params']) self.model = model self.model.summary() class CNNLSTMModel(ClassificationModel): __architect_name__ = 'CNNLSTMModel' __base_hyper_parameters__ = { 'conv_layer': { 'filters': 32, 'kernel_size': 3, 'padding': 'same', 'activation': 'relu' }, 'max_pool_layer': { 'pool_size': 2 }, 'lstm_layer': { 'units': 100 }, 'activation_layer': { 'activation': 'softmax' }, 'optimizer': { 'module': 'keras.optimizers', 'name': 'Adam', 'params': { 'lr': 1e-3, 'decay': 0.0 } }, 'compile_params': { 'loss': 'categorical_crossentropy', # 'optimizer': 'adam', 'metrics': ['accuracy'] } } def build_model(self): base_model = self.embedding.model conv_layer = Conv1D(**self.hyper_parameters['conv_layer'])(base_model.output) max_pool_layer = MaxPooling1D(**self.hyper_parameters['max_pool_layer'])(conv_layer) lstm_layer = LSTMLayer(**self.hyper_parameters['lstm_layer'])(max_pool_layer) dense_layer = Dense(len(self.label2idx), **self.hyper_parameters['activation_layer'])(lstm_layer) output_layers = [dense_layer] model = Model(base_model.inputs, output_layers) optimizer = getattr(eval(self.hyper_parameters['optimizer']['module']), self.hyper_parameters['optimizer']['name'])( **self.hyper_parameters['optimizer']['params']) model.compile(optimizer=optimizer, **self.hyper_parameters['compile_params']) self.model = model self.model.summary() class AVCNNModel(ClassificationModel): __architect_name__ = 'AVCNNModel' __base_hyper_parameters__ = { 'spatial_dropout': { 'rate': 0.25 }, 'conv_0': { 'filters': 300, 'kernel_size': 1, 'kernel_initializer': 'normal', 'padding': 'valid', 'activation': 'relu' }, 'conv_1': { 'filters': 300, 'kernel_size': 2, 'kernel_initializer': 'normal', 'padding': 'valid', 'activation': 'relu' }, 'conv_2': { 'filters': 300, 'kernel_size': 3, 'kernel_initializer': 'normal', 'padding': 'valid', 'activation': 'relu' }, 'conv_3': { 'filters': 300, 'kernel_size': 4, 'kernel_initializer': 'normal', 'padding': 'valid', 'activation': 'relu' }, # --- 'attn_0': {}, 'avg_0': {}, 'maxpool_0': {}, # --- 'maxpool_1': {}, 'attn_1': {}, 'avg_1': {}, # --- 'maxpool_2': {}, 'attn_2': {}, 'avg_2': {}, # --- 'maxpool_3': {}, 'attn_3': {}, 'avg_3': {}, # --- 'v0_col': { # 'mode': 'concat', 'axis': 1 }, 'v1_col': { # 'mode': 'concat', 'axis': 1 }, 'v2_col': { # 'mode': 'concat', 'axis': 1 }, 'merged_tensor': { # 'mode': 'concat', 'axis': 1 }, 'dropout': { 'rate': 0.7 }, 'dense': { 'units': 144, 'activation': 'relu' }, 'activation_layer': { 'activation': 'softmax' }, 'optimizer': { 'module': 'keras.optimizers', 'name': 'Adam', 'params': { 'lr': 1e-3, 'decay': 1e-7 }