For many machine learning algorithms, the final problem to be solved is often to minimize a function, which we usually call a loss function. The same is true in neural networks, where the loss function layer (CostLayer) and Optimizers come into being (...), where: Optimizers are used to minimize this loss It is worth mentioning that in the neural network, the loss can be understood in this way: it is the difference between the output obtained by inputting x and the real label y. How to define this gap and how to narrow this gap will involve a lot of mathematics. We will only talk about implementation here, the content of mathematics (with time) (that is, basically impossible) (hey) will be in the math series. Description. Thanks to the versatile tensorflow, which helps us define the loss function and Optimizers intimately, so we just need to package them. CostLayer First define a base class: This is equivalent to "stolen" the Layer's activation function into a loss function. The calculate function is used to directly calculate the loss, it is only used when analyzing the model's performance in a complex model, and can be temporarily ignored. To define the actual application of CostLayer, we take the most widely used CrossEntropy as an example: This is the built-in function of tensorflow. Optimizers This part of the package is made more nutritious, and most of the code is just for the interface with my own wheel. The most critical part is only two lines: Where self._opt is one of the Optimizers that tensorflow has defined for us. Its role is also very simple and rude: update each variable in the session to make the loss x move toward the minimum Above, the definition, function, and implementation of CostLayer and Optimizers are almost the same; together with the previous chapters, a complete, simpler neural network is fully implemented, and it supports the following functions: Custom activation function Arbitrarily stacked Layer Stacking duplicate structures by looping Evaluate the quality of the model by accuracy This is not a good model, but it has a basic prototype, and this step can be considered as an end. If you want to expand, the general process will be as follows: Record the results of the current training during the training to draw a curve like this: Let the model support the training of large-scale data, which includes several areas for improvement: We don't currently split the data into small batches to train our model; but when the amount of data is large, this kind of processing is indispensable. We are currently making predictions by throwing the entire data into the model to make it a pre-transfer algorithm. When the amount of data is large, this will cause the problem of insufficient memory. For this reason, we need to pre-deliver and make an integration at the end. We currently do not cross-validate, which makes our model easier to overfit. Although it is ok to let users divide the data by themselves, it is good practice to leave an interface. Last but not least, of course, we extend our model to support a CNN model. This is a huge pit, and let me slowly fill it... disposable vape pen Onlyrelx 1600puffs is portable and fashion disposable pod, disposable electronic cigarettes are trending featured vape pen for vapors as it's safety and easy to use. Disposable vape pod,disposable vape, wholesale vape,vape wholesale,vape pen manufacturer and supplier.disposable vape pen,disposable electronic cigarette,disposable ecigs pen,disposable ecigs stick,disposable e-cigs pen,disposable vape factory,disposable vape pod,disposable vape device,vape pen,vape stick, vape wholesale,wholesale vape,customized disposable vape pen,customized vape pen,OEM&ODM disposable ecigs pen,disposable electronic cigarette wholesale, wholesale disposable electronic cigarette,distribute vape pen,vape pen distribute,high quality vape pen,high quality vape pod,rechargeable disposable vape pen,refillable vape pen,refilling electronic cigarette,rechargeable disposable electronic cigarette,refillable vape pod,disposable refillable ecigs,disposable refilling e-cigs pen,refillable e-cigs pen Onlyrelx 1600puffs no ring,Disposable Vape Pen,Disposable Vaporizer, E-Cigarette, Vapes Pen,disposable pod,disposable bar Shenzhen Onlyrelx Technology Co.,Ltd , https://www.onlyrelxtech.com
