Tak-Wah Blog

持续学习语义理解框架ERNIE：https://aistudio.baidu.com/aistudio/projectdetail/1244949ERNIE/ERNIE 2.0 原理解析https://suixinblog.cn/2019/11/ernie.html关于ERNIE更详细的介绍，可以参考这两篇学术论文：ERNIE: Enhanced Representation through Knowledge Integration ERNIE 2.0: A Continual Pre-..

DatasetsThe reddit dataset should be downloaded from the following links and placed in directory./data. The details for Reddit Dataset can be foundhere.reddit.npzhttps://drive.google.com/open?id=19SphVl_Oe8SJ1r87Hr5a6znx3nJu1F2J reddit_adj.npz:htt...

官方链接：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/gather_cn.htmlpaddle.fluid.layers.gather(input,index,overwrite=True)根据索引index获取输入（input）的最外层维度的条目，并将它们拼接在一起。功能和tensorflow中的embedding_lookup相似。示例：import paddle.fl...

（1）bpr_loss：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/bpr_loss_cn.html（2）center_loss：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/center_loss_cn.html（3）dice_loss：https://www.paddlepaddle.org.cn/doc

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/expand_cn.htmlpaddle.fluid.layers.expand(x,expand_times,name=None)根据参数expand_times对输入x的各维度进行复制。通过参数expand_times来为x的每个维度设置复制次数。示例：import paddle.fluid as fluid...

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/dropout_cn.html示例：import paddle.fluid as fluidimport numpy as npx = fluid.data(name="x", shape=[-1, 4, 4], dtype="float32")droped = fluid.layers.dropout(x, dropout_prob.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/cross_entropy_cn.html示例：import paddle.fluid as fluidimport numpy as npclass_num = 7x = fluid.data(name='x', shape=[-1, 1, 10], dtype='float32')label = fluid.data(name=.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/cosine_decay_cn.htmlpaddle.fluid.layers.cosine_decay(learning_rate,step_each_epoch,epochs)使用 cosine decay 的衰减方式进行学习率调整。在训练模型时，建议一边进行训练一边降低学习率。 通过使用此方法，学习速率将通过如下cosine衰减策...

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/cos_sim_cn.html计算两个向量的余弦。示例：import paddle.fluid as fluidimport numpy as npx = fluid.data(name='x', shape=[3, 7], dtype='float32')y = fluid.data(name='y', shape=[1,..

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/conv2d_cn.html示例：import paddle.fluid as fluidimport numpy as npdata = fluid.data(name='data', shape=[-1, 3, 4, 4], dtype='float32')param_attr = fluid.ParamAttr(name='co.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/clip_cn.html示例：import paddle.fluid as fluidimport numpy as npinput = fluid.data( name='data', shape=[-1, 5], dtype='float32')reward = fluid.layers.clip(x=input, mi.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/ceil_cn.html示例：import paddle.fluid as fluidimport numpy as npinput_ceil = np.array([[-1.5, 6], [1, 15.6]])with fluid.dygraph.guard(): x = fluid.dygraph.to_variable.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/batch_norm_cn.html示例：import paddle.fluid as fluidimport numpy as np#x = fluid.layers.data(name='x', shape=[3, 7, 3, 7], dtype='float32', append_batch_size=False)x = flu.

官方文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/fc_cn.html#fc示例：import paddle.fluid as fluidimport numpy as np# 当输入为单个张量时data = fluid.data(name="data", shape=[1, 2, 3], dtype="float32")#data = fluid.layers.data(nam.

文档链接：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/argmax_cn.html示例：import paddle.fluid as fluidimport numpy as npx1 = fluid.layers.data(name="x1_", shape=[2, 3, 4], dtype="float32")print(x1.shape, "\n")out = fluid.l.

文档地址：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/concat_cn.html#concat示例：import paddle.fluid as fluidimport numpy as npx1 = fluid.layers.data(name="x1_", shape=[2, 3], dtype="float32")x2 = fluid.layers.data(name="x2_".

函数文档：https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn/flatten_cn.html#flatten示例如下：import paddle.fluid as fluidimport numpy as npx = fluid.layers.data(name="x", shape=[3, 2, 3], append_batch_size=False, dtype="float32")out .

见代码import paddle.fluid as fluidimport numpy as np# data, embeddingdata = fluid.layers.data(name="sequence", shape=[1], dtype="int64")emb_1 = fluid.layers.embedding(input=data, size=[12, 8])# data 赋值feed_data = x = np.array([[1], [2]])print("

直接见代码：import paddle.fluid as fluidimport numpy as npdata = fluid.layers.data(name='data', shape=[1], dtype='int64')np.random.seed(28)weight_data = np.random.random(size=(20, 8))print(weight_data)#加载用户自定义或预训练的词向量w_param_attrs = fluid.ParamAttr