Add CMBF model #217
Add CMBF model #217
Conversation
easy_rec/python/model/rank_model.py
Outdated
| @@ -300,7 +322,7 @@ def _build_distribute_metric_impl(self, | |||
| else: | |||
| raise ValueError('Wrong class number') | |||
| elif metric.WhichOneof('metric') == 'gauc': | |||
| assert loss_type == LossType.CLASSIFICATION | |||
| assert loss_type in [LossType.CLASSIFICATION, LossType.F1_REWEIGHTED_LOSS, LossType.PAIR_WISE_LOSS] | |||
There was a problem hiding this comment.
这个有没有更好的方式来实现，这样每次新加一个 loss 函数，都需要在每个 metric 下面都加一下，感觉有点繁琐
easy_rec/python/model/cmbf.py
Outdated
| attention_probs_dropout_prob=self._model_config.attention_probs_dropout_prob, | ||
| name='text_self_attention' | ||
| ) # shape: [batch_size, txt_seq_length, hidden_size] | ||
| print('txt_attention_fea:', txt_attention_fea.shape) |
There was a problem hiding this comment.
print 可以改成 logging 好像更合适一点
There was a problem hiding this comment.
print打印在stdout，比较清晰地查看；如果用logging在打印在stderr，stderr里面有很多系统输出，日志量非常大，不容易发现需要被注意的关键信息。
| @@ -27,7 +27,7 @@ def __init__(self, | |||
| labels, is_training) | |||
| self._loss_type = self._model_config.loss_type | |||
| self._num_class = self._model_config.num_class | |||
|
|
|||
| self._losses = self._model_config.losses | |||
There was a problem hiding this comment.
为什么不能用 self._losses 而要额外定义一个 self._losses
There was a problem hiding this comment.
losses是为了支持同时使用多个损失函数，而且可以为每个损失函数配置不同的相对权重。
| @@ -71,6 +71,8 @@ message FeatureConfig { | |||
|
|
|||
| // delimeter to separate sequence multi-values | |||
| optional string seq_multi_sep = 101; | |||
| // truncate sequence data to max_seq_len | |||
| optional uint32 max_seq_len = 102; | |||
There was a problem hiding this comment.
这个值是用来截断序列的，如果序列长度很大，则后续模型的计算量会很大。
|
CI PY3 Test Failed |
|
CI Test Failed |
|
CI PY3 Test Failed |
|
CI Test Failed |
|
CI PY3 Test Failed |
|
CI Test Failed |
|
CI Test Passed |
github-actions
bot
added
ci_test_passed
|
CI Test Passed |
github-actions
bot
added
ci_test_passed
|
CI PY3 Test Passed |
|
CI PY3 Test Failed |
|
CI Test Passed |
github-actions
bot
added
ci_test_passed
|
CI PY3 Test Passed |
|
CI PY3 Test Passed |
|
CI Test Passed |
github-actions
bot
added
ci_test_passed
| # | ||
| # @unittest.skipIf(gl is None, 'graphlearn is not installed') | ||
| # def test_dssm_neg_sampler_v2(self): | ||
| # self._success = test_utils.test_single_train_eval( |
There was a problem hiding this comment.
these unit tests should not be removed.
No description provided.