您当前的位置:首页 > IT编程 > Keras
| C语言 | Java | VB | VC | python | Android | TensorFlow | C++ | oracle | 学术与代码 | cnn卷积神经网络 | gnn | 图像修复 | Keras | 数据集 | Neo4j | 自然语言处理 | 深度学习 | 医学CAD | 医学影像 | 超参数 | pointnet | pytorch | 异常检测 | Transformers | 情感分类 | 知识图谱 |

自学教程:Python layers.Softmax方法代码示例

51自学网 2020-12-01 11:09:09
  Keras
这篇教程Python layers.Softmax方法代码示例写得很实用,希望能帮到您。

本文整理汇总了Python中keras.layers.Softmax方法的典型用法代码示例。如果您正苦于以下问题:Python layers.Softmax方法的具体用法?Python layers.Softmax怎么用?Python layers.Softmax使用的例子?那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在模块keras.layers的用法示例。

在下文中一共展示了layers.Softmax方法的8个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于我们的系统推荐出更棒的Python代码示例。

示例1: __output

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def __output(self, dec_output):        output_dropout_layer = Dropout(self.output_dropout)        output_layer = Conv1D(self.tgt_vocab_size + 1,                              kernel_size=1,                              activation=gelu,                              kernel_regularizer=regularizers.l2(self.l2_reg_penalty),                              name='output_layer')        output_softmax_layer = Softmax(name="word_predictions")        if self.use_crf:            return output_layer(output_dropout_layer(dec_output))        else:            return output_softmax_layer(output_layer(output_dropout_layer(dec_output))) 
开发者ID:GlassyWing,项目名称:transformer-word-segmenter,代码行数:18,代码来源:__init__.py


示例2: create_model

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def create_model(input_shape: tuple, nb_classes: int, init_with_imagenet: bool = False, learning_rate: float = 0.01):    weights = None    if init_with_imagenet:        weights = "imagenet"    model = VGG16(input_shape=input_shape,                  classes=nb_classes,                  weights=weights,                  include_top=False)    # "Shallow" VGG for Cifar10    x = model.get_layer('block3_pool').output    x = layers.Flatten(name='Flatten')(x)    x = layers.Dense(512, activation='relu')(x)    x = layers.Dense(nb_classes)(x)    x = layers.Softmax()(x)    model = models.Model(model.input, x)    loss = losses.categorical_crossentropy    optimizer = optimizers.SGD(lr=learning_rate, decay=0.99)    model.compile(optimizer, loss, metrics=["accuracy"])    return model 
开发者ID:gaborvecsei,项目名称:Federated-Learning-Mini-Framework,代码行数:24,代码来源:models.py


示例3: build_resnet_generator

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def build_resnet_generator(input_shape, n_filters, n_residual_blocks,                           seq_len, vocabulary_size):    inputs = Input(shape=input_shape)    # Dense 1: 1 x seq_len x n_filters    x = Dense(1 * seq_len * n_filters, input_shape=input_shape)(inputs)    x = Reshape((1, seq_len, n_filters))(x)    # ResNet blocks    x = resnet_block(x, n_residual_blocks, n_filters)    # Output layer    x = Conv2D(filters=vocabulary_size, kernel_size=1, padding='same')(x)    x = Softmax(axis=3)(x)    # create model graph    model = Model(inputs=inputs, outputs=x, name='Generator')    print("/nGenerator ResNet")    model.summary()    return model 
开发者ID:PacktPublishing,项目名称:Hands-On-Generative-Adversarial-Networks-with-Keras,代码行数:23,代码来源:models.py


示例4: __call__

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def __call__(self, q, k, v, attn_mask=None, scale=1.0):        """        :param q: Queries 张量,形状为[N, T_q, D_q]        :param k: Keys 张量,形状为[N, T_k, D_k]        :param v: Values 张量,形状为[N, T_v, D_v]        :param attn_mask: 注意力掩码,形状为[N, T_q, T_k]        :param scale: 缩放因子,浮点标量        :return: 上下文张量和注意力张量        """        attention = Lambda(lambda x: K.batch_dot(x[0], x[1], axes=(2, 2)) * scale)([q, k])  # [N, T_q, T_k]        if attn_mask is not None:            # 为需要掩码的地方设置一个负无穷,softmax之后就会趋近于0            attention = Lambda(lambda x: (-1e+10) * (1 - x[0]) + x[1])([attn_mask, attention])        attention = Softmax(axis=-1)(attention)        attention = Dropout(self.attention_dropout)(attention)  # [N, T_q, T_k]        context = Lambda(lambda x: K.batch_dot(x[0], x[1], axes=(2, 1)))([attention, v])  # [N, T_q, D_q]        return context, attention 
开发者ID:GlassyWing,项目名称:transformer-keras,代码行数:21,代码来源:core.py


示例5: test_softmax

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def test_softmax():    for axis in [1, -1]:        layer_test(layers.Softmax, kwargs={'axis': axis},                   input_shape=(2, 3, 4)) 
开发者ID:hello-sea,项目名称:DeepLearning_Wavelet-LSTM,代码行数:6,代码来源:advanced_activations_test.py


示例6: __init__

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def __init__(self,                 src_vocab_size,                 src_max_len,                 tgt_vocab_size,                 tgt_max_len,                 optimizer=Adam(lr=1e-3),                 num_layers=6,                 model_dim=512,                 num_heads=8,                 ffn_dim=2048,                 dropout=0.2,                 src_tokenizer=None,                 tgt_tokenizer=None,                 weights_path=None):        self.optimizer = optimizer        self.src_max_len = src_max_len        self.tgt_max_len = tgt_max_len        self.src_vocab_size = src_vocab_size        self.tgt_vocab_size = tgt_vocab_size        self.model_dim = model_dim        self.num_layers = num_layers        self.num_heads = num_heads        self.ffn_dim = ffn_dim        self.dropout = dropout        self.decode_model = None  # used in beam_search        self.encode_model = None  # used in beam_search        self.src_tokenizer = src_tokenizer        self.tgt_tokenizer = tgt_tokenizer        self.encoder = Encoder(src_vocab_size, src_max_len, num_layers, model_dim,                               num_heads, ffn_dim, dropout)        self.decoder = Decoder(tgt_vocab_size, tgt_max_len, num_layers, model_dim,                               num_heads, ffn_dim, dropout)        self.linear = Dense(tgt_vocab_size + 1, use_bias=False)        self.softmax = Softmax(axis=2)        self.pred_model, self.model = self.__build_model()        if weights_path is not None:            self.model.load_weights(weights_path) 
开发者ID:GlassyWing,项目名称:transformer-keras,代码行数:43,代码来源:core.py


示例7: universal_transformer_gpt_model

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def universal_transformer_gpt_model(        max_seq_length: int, vocabulary_size: int,        word_embedding_size: int, transformer_depth: int,        num_heads: int, transformer_dropout: float = 0.1,        embedding_dropout: float = 0.6,        l2_reg_penalty: float = 1e-6,        confidence_penalty_weight: float = 0.1):    """    A model which is similar to the one described by OpenAI in paper    "Improving Language Understanding by Generative Pre-Training", except    that it relies L2 regularization of the word embedding matrix    (instead of the dropout), and uses Universal Transformer architecture.    """    word_ids = Input(shape=(max_seq_length,), dtype='int32', name='word_ids')    l2_regularizer = (regularizers.l2(l2_reg_penalty) if l2_reg_penalty                      else None)    embedding_layer = ReusableEmbedding(        vocabulary_size, word_embedding_size,        input_length=max_seq_length,        name='bpe_embeddings',        # Regularization is based on paper "A Comparative Study on        # Regularization Strategies for Embedding-based Neural Networks"        # https://arxiv.org/pdf/1508.03721.pdf        embeddings_regularizer=l2_regularizer)    output_layer = TiedOutputEmbedding(        projection_regularizer=l2_regularizer,        projection_dropout=embedding_dropout,        name='word_prediction_logits')    coordinate_embedding_layer = TransformerCoordinateEmbedding(        transformer_depth,        name='coordinate_embedding')    transformer_act_layer = TransformerACT(name='adaptive_computation_time')    transformer_block = TransformerBlock(        name='transformer', num_heads=num_heads,        residual_dropout=transformer_dropout,        attention_dropout=transformer_dropout,        use_masking=True, vanilla_wiring=False)    output_softmax_layer = Softmax(name='word_predictions')    next_step_input, embedding_matrix = embedding_layer(word_ids)    act_output = next_step_input    for i in range(transformer_depth):        next_step_input = coordinate_embedding_layer(next_step_input, step=i)        next_step_input = transformer_block(next_step_input)        next_step_input, act_output = transformer_act_layer(next_step_input)    transformer_act_layer.finalize()    next_step_input = act_output    word_predictions = output_softmax_layer(        output_layer([next_step_input, embedding_matrix]))    model = Model(inputs=[word_ids], outputs=[word_predictions])    # Penalty for confidence of the output distribution, as described in    # "Regularizing Neural Networks by Penalizing Confident    # Output Distributions" (https://arxiv.org/abs/1701.06548)    confidence_penalty = K.mean(        confidence_penalty_weight *        K.sum(word_predictions * K.log(word_predictions), axis=-1))    model.add_loss(confidence_penalty)    return model 
开发者ID:kpot,项目名称:keras-transformer,代码行数:62,代码来源:models.py


示例8: vanilla_transformer_gpt_model

# 需要导入模块: from keras import layers [as 别名]# 或者: from keras.layers import Softmax [as 别名]def vanilla_transformer_gpt_model(        max_seq_length: int, vocabulary_size: int,        word_embedding_size: int, transformer_depth: int,        num_heads: int, transformer_dropout: float = 0.1,        embedding_dropout: float = 0.6,        l2_reg_penalty: float = 1e-6,        confidence_penalty_weight: float = 0.1):    """    A model which is almost identical to the one described by OpenAI in paper    "Improving Language Understanding by Generative Pre-Training", except    that it uses L2 regularization of the word embedding matrix,    instead of the dropout.    """    word_ids = Input(shape=(max_seq_length,), dtype='int32', name='word_ids')    l2_regularizer = (regularizers.l2(l2_reg_penalty) if l2_reg_penalty                      else None)    embedding_layer = ReusableEmbedding(        vocabulary_size, word_embedding_size,        input_length=max_seq_length,        name='bpe_embeddings',        # Regularization is based on paper "A Comparative Study on        # Regularization Strategies for Embedding-based Neural Networks"        # https://arxiv.org/pdf/1508.03721.pdf        embeddings_regularizer=l2_regularizer)    output_layer = TiedOutputEmbedding(        projection_regularizer=l2_regularizer,        projection_dropout=embedding_dropout,        name='word_prediction_logits')    coordinate_embedding_layer = TransformerCoordinateEmbedding(        1,        name='coordinate_embedding')    output_softmax_layer = Softmax(name='word_predictions')    next_step_input, embedding_matrix = embedding_layer(word_ids)    next_step_input = coordinate_embedding_layer(next_step_input, step=0)    for i in range(transformer_depth):        next_step_input = (            TransformerBlock(                name='transformer' + str(i), num_heads=num_heads,                residual_dropout=transformer_dropout,                attention_dropout=transformer_dropout,                use_masking=True,                vanilla_wiring=True)            (next_step_input))    word_predictions = output_softmax_layer(        output_layer([next_step_input, embedding_matrix]))    model = Model(inputs=[word_ids], outputs=[word_predictions])    # Penalty for confidence of the output distribution, as described in    # "Regularizing Neural Networks by Penalizing Confident    # Output Distributions" (https://arxiv.org/abs/1701.06548)    confidence_penalty = K.mean(        confidence_penalty_weight *        K.sum(word_predictions * K.log(word_predictions), axis=-1))    model.add_loss(confidence_penalty)    return model 
开发者ID:kpot,项目名称:keras-transformer,代码行数:59,代码来源:models.py


万事OK自学网:51自学网_软件自学网_CAD自学网自学excel、自学PS、自学CAD、自学C语言、自学css3实例,是一个通过网络自主学习工作技能的自学平台,网友喜欢的软件自学网站。