hi主要是没太多时间用来思考，所以很多东西都不是太清楚，只会装逼，这样不好，要从理论/代码上理解到底说的是什么事情，这是科学问题：是什么和为什么？作为搬砖工还是有必要深入了解，不然真的只能搬砖了。可在我的主页搜索相关博文，不再挨个附上。【面试明明是技术面，却总是在问科学的问题，这很扯淡】
For Recommendation in Deep learning QQ Group 102948747
For Visual in deep learning QQ Group 629530787
不接受这个网页的私聊/私信！！！
今年的目标没有实现——访客50万，总排名5000，明年三月份要实现这个东西。
之前实现了faiss召回，领导又问了，为啥采用IP的方式？其意义在哪里？卧槽，这是真的防不胜防啊，ytb中也是如此啊，但真的没有细致了解这玩意，这里还是很真探讨下这些问题。attention及GRU的作用是什么？为啥用GRU而不是LSTM，RNN？卧槽，随便一问就是面试题中的致命点，既是送分题，也是送命题。
【20210103续】
只要糖放得多，你就不会觉得生活有多苦，几块钱的事，非要百万来解决，何必为难自己。
试了下RNN，结果稍微差点，但绝对差不到一个百分点（HR一个为0.262，一个为0.268，MAP0.0506，一个为0.0532，好的结果是GRU的），昨晚也试了LSTM，但直接替换函数不对，卧槽，再改改吧，报错如下：
Traceback (most recent call last):
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 854, in __call__
outputs = call_fn(cast_inputs, *args, **kwargs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/autograph/impl/api.py", line 237, in wrapper
raise e.ag_error_metadata.to_exception(e)
tensorflow.python.framework.errors_impl.OperatorNotAllowedInGraphError: in converted code:
relative to /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python:
ops/rnn_cell_impl.py:1017 call
(c_prev, m_prev) = state
framework/ops.py:547 __iter__
self._disallow_iteration()
framework/ops.py:543 _disallow_iteration
self._disallow_in_graph_mode("iterating over `tf.Tensor`")
framework/ops.py:523 _disallow_in_graph_mode
" this function with @tf.function.".format(task))
OperatorNotAllowedInGraphError: iterating over `tf.Tensor` is not allowed in Graph execution. Use Eager execution or decorate this function with @tf.function.
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/data/logs/xulm1/myconda/lib/python3.7/multiprocessing/process.py", line 297, in _bootstrap
self.run()
File "/data/logs/xulm1/myconda/lib/python3.7/multiprocessing/process.py", line 99, in run
self._target(*self._args, **self._kwargs)
File "video_update_rec_items.py", line 801, in run_train
model = Model(node_count+1, checkpoint_path, **kwargs)
File "video_update_rec_items.py", line 304, in __init__
self.loss, self.session_state, self.logits = self.__forward()
File "video_update_rec_items.py", line 345, in __forward
node_state = self.ggnn(self.adj_in, self.adj_out, node_state, nw)
File "video_update_rec_items.py", line 216, in __call__
_, state = self.gru_cell(tf.reshape(gru_in, [-1, 2*self.hidden_size]), tf.reshape(state, [-1, self.hidden_size]))
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/ops/rnn_cell_impl.py", line 386, in __call__
self, inputs, state, scope=scope, *args, **kwargs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/layers/base.py", line 548, in __call__
outputs = super(Layer, self).__call__(inputs, *args, **kwargs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 861, in __call__
str(e) + '\n"""')
TypeError: You are attempting to use Python control flow in a layer that was not declared to be dynamic. Pass `dynamic=True` to the class constructor.我以为直接按照提示改就行了，卧槽。还是不行啊。咋整啊，
>>> import tensorflow as tf
>>> tf.compat.v1.nn.rnn_cell.LSTMCell(32,dynamic=True)
WARNING:tensorflow:From <stdin>:1: LSTMCell.__init__ (from tensorflow.python.ops.rnn_cell_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This class is equivalent as tf.keras.layers.LSTMCell, and will be replaced by that in Tensorflow 2.0.
<tensorflow.python.ops.rnn_cell_impl.LSTMCell object at 0x7fdacf531550>
>>> quit()
报错如下：这下好了，更没法改了
Traceback (most recent call last):
File "/data/logs/xulm1/myconda/lib/python3.7/multiprocessing/process.py", line 297, in _bootstrap
self.run()
File "/data/logs/xulm1/myconda/lib/python3.7/multiprocessing/process.py", line 99, in run
self._target(*self._args, **self._kwargs)
File "video_update_rec_items.py", line 801, in run_train
model = Model(node_count+1, checkpoint_path, **kwargs)
File "video_update_rec_items.py", line 304, in __init__
self.loss, self.session_state, self.logits = self.__forward()
File "video_update_rec_items.py", line 345, in __forward
node_state = self.ggnn(self.adj_in, self.adj_out, node_state, nw)
File "video_update_rec_items.py", line 216, in __call__
_, state = self.gru_cell(tf.reshape(gru_in, [-1, 2*self.hidden_size]), tf.reshape(state, [-1, self.hidden_size]))
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/ops/rnn_cell_impl.py", line 386, in __call__
self, inputs, state, scope=scope, *args, **kwargs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/layers/base.py", line 548, in __call__
outputs = super(Layer, self).__call__(inputs, *args, **kwargs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 869, in __call__
outputs = self._symbolic_call(inputs)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 2158, in _symbolic_call
output_shapes = self.compute_output_shape(input_shapes)
File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/base_layer.py", line 646, in compute_output_shape
raise NotImplementedError
NotImplementedError还是搜索下LSTM与GRU的基本用法吧，看看是不是有所区别。好难哦，我看LSTM基本上都是和dynamic_rnn一起使用的，而我这里没有，加了这个就更难改了。
我看还是算了，这个就当实现了吧，问就是效果差点，，，，小明哥你不能这样，不讲武德。。。我特么源码给你，要不你帮我改成LSTM的，我跪谢。
实在是能力有限，改就伤筋动骨了，不好整。
【问题2：为啥用IP求内积/相似度/距离？两个embedding求IP的意义是什么？】（这才是领导问的，小声bb，其他的是我自己发问的）
这其实是一个问题，下面从头来过
将所有的会话图喂到ggnn中，得到所有节点的embedding，然后通过聚合得到会话的全局embedding，采用的是self-attention机制更好的表达。会话局部的embedding可简单直接认为就是最后一个点击的item的embedding。两个会话的embedding通过concat后线性转换得到最后会话表达。这是会话embedding的来源。
为给每一个item计算score，只需将会话的embedding与所有的item的embedding相乘，然后取softmax就得到了下次点击item出现的概率。这也是训练和推理阶段相同的。如果你还不懂，不妨留个坑，待我看完ytb的详解后再次解答。
其实也无需再说，在ytb中得到的user和item的embedding，为何他们要相乘再softmax呢？这是一样的道理，只不过这里是会话的表达，你可以将会话表达看做是用户的兴趣/偏好表达。
拜拜，愿我们终有重逢之时，而你还记得我们曾经讨论的话题。