Multi-model fusion method and device, electronic equipment and computer readable storage medium
A fusion method and model fusion technology, which is applied in the field of machine learning and can solve problems such as being unrobust, not suitable for continuous training of streaming data, and not suitable for semi-supervised learning methods in its basic form.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
example 1
[0101]
[0102] image 3 It is the model structure diagram of the fusion model obtained by the mean fusion method. like image 3 As shown, in the mean fusion method, the training results of the two sub-models trained in the model pre-training are used for model fusion. Specifically, use the trained TabTransformer and LGB on the test dataset (x Test ,y Test ) to make predictions and get the prediction results p respectively Tab_Test and p LGB_Test , and average the prediction results of the two, that is, p=(p Tab_Test +p LGB_Test ) / 2, as the prediction result of the fusion model.
example 2
[0103]
[0104] Figure 4 It is the model structure diagram of the fusion model obtained by the residual fusion method. like Figure 4 As shown, in the residual fusion method, the training data set (x T ,y T ), validation dataset (x V ,y V ) and the test dataset (x Test ,y Test ) to predict respectively, and get the prediction result p Tab_T , p Tab_V and p Tab_Test , and calculate the prediction result p respectively Tab_T , p Tab_V The residual between the predicted target y, i.e. get respectively and Then, replace the prediction targets in the training dataset and validation dataset with these residuals to get the replaced training dataset and the replaced validation dataset Train the LGB model with the replaced training dataset and validation dataset, and use the trained LGB model on the test dataset (x Test ,y Test ) to make predictions and get the prediction result p LGB_Test , and finally, with p=(p Tab_Test +p LGB_Test ) as the prediction res...
example 3
[0105]
[0106] Figure 5 It is the model structure diagram of the fusion model obtained by the neural network graft fusion method. like Figure 5 As shown, in the neural network graft fusion, the training data set (x T ,y T ) and the validation dataset (x V ,y V ) to predict and get the test result p LGB_T and p LGB_V . Then, insert the obtained test result into the network operation of TabTransformer. Specifically, because the test result belongs to numerical data, it is different from the original x cont The c consecutive features of , together form c+1 features, and are embedded with the context {h in the concatenation layer 1 ,…,h m } are concatenated together to form a vector of dimension (d×m+c+1), which is then fed into the represents the MLP layer to predict the target y. Here, the loss function is:
[0107]
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More