LLLLLLoki

感谢您的工作，我在使用您的代码做NER 时出现了一下问题： 2020-08-03 19:11:21.629155: F tensorflow/lite/toco/graph_transformations/propagate_fixed_sizes.cc:1813] Check failed: axis < input_shape.dimensions_count() (62 vs. 3) Fatal Python error: Aborted Current thread 0x00007f0435b05700 (most recent call first): File "/home/lxj/anaconda3/envs/nlp-gpu1.14/lib/python3.7/site-packages/tensorflow/lite/toco/python/toco_from_protos.py", line 33 in...

请问 您在loss中加入 with tf.variable_scope("losses_avg"): loss_moving_avg = tf.get_variable("training_loss", initializer=100.0, trainable=False) lm = loss_moving_avg.assign(loss_moving_avg * 0.9 + loss * 0.1) tf.summary.scalar("loss_avg/total_loss", lm) with tf.control_dependencies([lm]): loss = tf.identity(loss) 为什么要加入一个losses_avg

在FFN中的问题

1

论文中我看到的FFN是FFN（X） = ReLU(XW1)W2,为什么在代码中的_ffn_layer 中的linear 函数里会有tf.nn.convolution？只有_linear_2d 没有卷积函数 其他的都有

cypher = "match p=shortestPath((n:Entity{name:"aaa"})-[*]-(m:Entity{name:"bbb"})) return p" g = Graph(...,...) PATH = g.run(cypher).to_series() How do I get the name attribute of the relationship in the PATH？

DiceLoss 这个公式写的对吗，怎么理解呢

1

def forward(self,input, target): ''' input: [N, C] target: [N, ] ''' prob = torch.softmax(input, dim=1) prob = torch.gather(prob, dim=1, index=target.unsqueeze(1)) dsc_i = 1 - ((1 - prob) * prob) /...

I trained a model myself. How do I test it online in real time

6

I trained a model myself. How do I test it online in real time？Should I rewrite a file like mobilenet_v2_tsm.py

How to train uni-direction TSM ? for example？？

2

hi ,Thank you very much for uploading the code of TSM. Now I want to trainuni-direction TSM . May I modify the code in this way def shift(x, n_segment, fold_div=3,...

请问flip， channel_shuffle， random_noise， random_brightness，random_contrast，random_saturation 等 为什么没有加入训练呢？

why query is ahead of context? inputs_ids = [CLS_ID] + [EOS_ID] + query_ids + [SEP_ID] + content_ids In theory it should be : inputs_ids = [CLS_ID] + [EOS_ID] + content_ids...