Keras Implementation of Structured data learning with TabTransformer
This repo contains the trained model of Structured data learning with TabTransformer.
The full credit goes to: Khalid Salama
Spaces Link:
Model summary:
- The trained model uses self-attention based Transformers structure following by multiple feed forward layers in order to serve supervised and semi-supervised learning.
- The model’s inputs can contain both numerical and categorical features.
- All the categorical features will be encoded into embedding vector with the same number of embedding dimensions, before adding (point-wise) with each other and feeding into a stack of Transformer blocks.
- The contextual embeddings of the categorical features after the final Transformer layer, are concatenated with the input numerical features, and fed into a final MLP block.
- A SoftMax function is applied at the end of the model.
Intended uses & limitations:
- This model can be used for both supervised and semi-supervised tasks on tabular data.
Training and evaluation data:
- This model was trained using the United States Census Income Dataset provided by the UC Irvine Machine Learning Repository. The task of the dataset is to predict whether a person is likely to be making over USD 50,000 a year (binary classification).
- The dataset consists of 14 input features: 5 numerical features and 9 categorical features.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- optimizer: ‘AdamW’
- learning_rate: 0.001
- weight decay: 1e-04
- loss: ‘sparse_categorical_crossentropy’
- beta_1: 0.9
- beta_2: 0.999
- epsilon: 1e-07
- epochs: 50
- batch_size: 16
- training_precision: float32
Training Metrics
Model history needed
Model Plot
View Model Plot
收录说明:
1、本网页并非 keras-io/tab_transformer 官网网址页面,此页面内容编录于互联网,只作展示之用;2、如果有与 keras-io/tab_transformer 相关业务事宜,请访问其网站并获取联系方式;3、本站与 keras-io/tab_transformer 无任何关系,对于 keras-io/tab_transformer 网站中的信息,请用户谨慎辨识其真伪。4、本站收录 keras-io/tab_transformer 时,此站内容访问正常,如遇跳转非法网站,有可能此网站被非法入侵或者已更换新网址,导致旧网址被非法使用,5、如果你是网站站长或者负责人,不想被收录请邮件删除:i-hu#Foxmail.com (#换@)
前往AI网址导航
2、本站所有文章、图片、资源等如果未标明原创,均为收集自互联网公开资源;分享的图片、资源、视频等,出镜模特均为成年女性正常写真内容,版权归原作者所有,仅作为个人学习、研究以及欣赏!如有涉及下载请24小时内删除;
3、如果您发现本站上有侵犯您的权益的作品,请与我们取得联系,我们会及时修改、删除并致以最深的歉意。邮箱: i-hu#(#换@)foxmail.com