DeepMatch icon indicating copy to clipboard operation
DeepMatch copied to clipboard

A deep matching model library for recommendations & advertising. It's easy to train models and to export representation vectors which can be used for ANN search.

Results 44 DeepMatch issues
Sort by recently updated
recently updated
newest added

各位大佬们,有一点不太理解的,当我把negsample设为0,这是相当于没有负样本给模型学习吗?如果是,为什么模型可以在没有负样本的情况下训练呢?如果不是,负样本是在哪个步骤生成的呢? P.S. 我看源码DSSM应该是没用上NegativeSampler?我试过把num_sampled随意设成负数或者是字符串也不影响模型的运行,比如这样: `sampler_config = NegativeSampler('inbatch',num_sampled='dfasdfas',item_name="item_id",item_count=item_count)` `model = DSSM(user_feature_columns, item_feature_columns,user_dnn_hidden_units=(128,64, embedding_dim), item_dnn_hidden_units=(64, embedding_dim,),loss_type='softmax',sampler_config=sampler_config) ` 所以顺便想问问,如果想在双塔模型生成in-batch的负样本,大佬们有什么好方法吗? 先谢谢大家了!

question

**Describe the question(问题描述)** 使用下面命令训练并保存模型,报ValueError: Unable to save the object ListWrapper([, ]) (a list wrapper constructed to track trackable TensorFlow objects). A list element was replaced (__setitem__, __setslice__), deleted (__delitem__, __delslice__),...

question

YoutubeDNN导出item侧模型错误,提示 raise AssertionError( AssertionError: Tried to export a function which references untracked resource Tensor("lambda_1/l2_normalize:0", shape=(209, 16), dtype=float32). TensorFlow objects (e.g. tf.Variable) captured by functions must be tracked by assigning them...

**Describe the bug(问题描述)** 当对DSSM模型设置sample_weight会报错,其中sample_weight是按照格式,和label一样大小的一个numpy的一维数组 **To Reproduce(复现步骤)** 运行代码: history = model.fit(train_model_input, train_label, batch_size=256, epochs=4, verbose=1, validation_split=0.0, sample_weight = sample_weights) **Operating environment(运行环境):** - python version [3.7] - tensorflow version [1.15.0] - deepmatch...

SDM item_feature 只支持一个item_id特征吗?

question

inner product是通过tf.reduce_sum实现的,reduce时候没有指定维度,建议默认axis=-1 当前0.3.1版本dssm模型无法构建,按如下方式修改后可以正常训练。 def inner_product(x, y, temperature=1.0, axis=-1): return Lambda(lambda x: tf.reduce_sum(tf.multiply(x[0], x[1]), axis) / temperature)([x, y])

Bumps [tensorflow](https://github.com/tensorflow/tensorflow) from 2.6.2 to 2.11.1. Release notes Sourced from tensorflow's releases. TensorFlow 2.11.1 Release 2.11.1 Note: TensorFlow 2.10 was the last TensorFlow release that supported GPU on native-Windows. Starting...

dependencies

update parameters with axis and keepdims

Please refer to the [FAQ](https://deepmatch.readthedocs.io/en/latest/FAQ.html) in doc and search for the [related issues](https://github.com/shenweichen/DeepMatch/issues) before you ask the question. **Describe the question(问题描述)** 跑完sdm模型,想导出用户塔用作实时召回,savemodel格式,失败,显示缺失state **Additional context** 日志如下: `WARNING:tensorflow:Skipping full serialization of Keras...

question

我的youtubednn训练很慢, 1024一个batch要花1分钟来训练, 也就是说500万行数据要2,3天, 同时显存占用率很高但是使用率低。 可能是数据读取比较慢但是测过数据读取时间应该不是这方面原因, 同样的数据读取格式DSSM只需要半小时, 差几十倍速度。 会不会是和我的mid size有关, 我的mid vocabulary size是1000万, 负采样num_sampled是500。 理论上SampledSoftmaxLoss的话和总的mid vocabulary size没有太大关系吧, 我也试过num_sampled=5,时间上没有变化。 不知道有人碰到过相似情况吗,谢谢。

question