Purpose
Understand the structure of fully connected neural networks through experiments, and apply fully connected networks to handle classification and regression tasks.
Experimental task
1) Primary experiments: ①Handwritten image recognition experiment; ②FashionMnist image classification experiment; ③Car mileage prediction experiment.
2) Intermediate experiment: ① Comparative experiment of iris flower classification task.
Experimental procedure
1. Handwritten image recognition experiment
1) Process the required data set, here the MNIST data set is used.
2) Define a network, here we use the LeNet network.
3) Define the loss function and optimizer.
4) Load the data set and perform training. After the training is completed, view the results and save the model file.
5) Load the saved model and perform inference.
6) Verify the model, load the test data set and the trained model, and verify the accuracy of the results.
2. Comparative experiment of iris flower classification task.
The optimizer is a very important structure in the training process of the neural network. It is precisely because of the existence of the optimizer that the network that initializes the parameters learns the optimal parameters that conform to the characteristics of the training set data step by step. This experiment mainly introduces how to use MindSpore to conduct optimization experiments. By analyzing the results of no optimizer, SGD optimizer, Momentum optimizer, and Adam optimizer models under the extreme point and classification tasks, the effects of different optimizers are obtained.
Experimental results
