v0.8.1

Change Log

Major Features and Improvements

• Improve the reproducibility of models

Bug Fixes and Other Changes

1. Fix ValueError: The name "seq_length" is used 2 times in the model. All layer names should be unique. when running run_dien.py in tf-2.x #186

2. Fix ImportError: cannot import name 'create_embedding_matrix when running
   from deepctr.inputs import create_embedding_matrix
   #252

3. Fix ValueError: The initial value's shape ((27, 4)) is not compatible with the explicitly supplied shape argument ((27, 1)). when running run_classification_criteo.py in tf-2.3.0 #259

v0.8.0

Change Log 中文版本更新说明

Major Features and Improvements

• Support Tensorflow Estimator for large scale data and distributed training.
  • example: Estimator with TFRecord
  • example: Estimator with Pandas DataFrame
  • Estimator Models API
• Support different initializers for different embedding weights by settting embeddings_initializer of SparseFeat
• Support using pretrained weights to initialize embedding weights and frozen embedding weights. example
• Add new model FwFM

Bug Fixes and Other Changes

API changes

• The followings are moved from deepctr.inputs to deepctr.feature_column
  • Class:SparseFeat,VarLenSparseFeat,DenseFeat
  • Function:get_feature_names,build_input_features,get_linear_logit,input_from_feature_columns
  • Variable:DEFAULT_GROUP_NAME
• combined_dnn_input is moved from deepctr.inputs to deepctr.layers.utils
• SparseFeat add embeddings_initializer and trainable arguments。

v0.7.5

Change Log

Bug Fixes and Other Changes

• Fix numerical instability in LayerNormalization

API changes

• Remove embedding_size parameter from PNN model

v0.7.4

Change Log

Major Features and Improvements

• Add FLEN Chinese Introducton FLEN: 一种时空高效的利用场信息缓解梯度耦合的大规模CTR预测模型
  

v0.7.3

Change Log

Bug Fixes and Other Changes

Fix the inconsistency of prediction results when the model is loaded with trained weights.

v0.7.2

Change Log

• fix bug in get_varlen_pooling_list
• fix bug in SequencePoolingLayer when mode='max'
• Add alias ONN for model NFFM

v0.7.1

Change Log 中文版本更新说明

Major Features and Improvements

• Simplify VarLenSparseFeat,support setting weight_normalization. #163

Bug Fixes and Other Changes

• Fix problem of embedding size of SparseFeat in linear_feature_columns. #178

API changes

deepctr.layers.sequence.WeightedSequenceLayer

default value of weight_normalization in WeightedSequenceLayer is set to True。

• previous：deepctr.layers.sequence.WeightedSequenceLayer(weight_normalization=False, supports_masking=False)

• now：deepctr.layers.sequence.WeightedSequenceLayer(weight_normalization=True, supports_masking=False)

deepctr.inputs.VarLenSparseFeat

Parameter list of VarLenSparseFeat consists of a instance of SparseFeat and parameters related to sequence.

• previous：VarLenSparseFeat(name, maxlen, vocabulary_size, embedding_dim=4, combiner="mean", use_hash=False, dtype="float32", length_name=None, weight_name=None, embedding_name=None, group_name=DEFAULT_GROUP_NAME)

• now：VarLenSparseFeat(sparsefeat, maxlen, combiner="mean", length_name=None, weight_name=None, weight_norm=True)

v0.7.0

Change Log

Major Features and Improvements

• Refactor feature columns.
• Different features can use different embedding_dim
• group-wise interaction is available by setting group_name.
• add linear part to some models

API changes

• embedding_size parameter of models is removed.Now we must set embedding_dim(default 4) in SparseFeat or VarLenSparseFeat .

v0.6.3

Change Log

Major Features and Improvements

Add WeightedSequenceLayer and support weighted sequence feature input.

Bug Fixes and Other Changes

fix bug of duplicated bias term in models.#150

v0.6.2

Major Features and Improvements

• Simplify input logic,examples