Completed today
For the virtual completion of the experiment samples.
That accounted for samples with different fictitious horizontal axis, and the accuracy of the comparison loss if there is improvement.
Results: There is no law at all.
Internet continue to look for papers.
Fortunately, on which the two articles of accuracy is not high, or even can be said to be very low, it gave me great information and inspiration.
Tomorrow target
Continue to do experiments, generalization reinspection neural network adequacy, today seems to have found some problems in the sample.
This experiment is a single-input single-output. That input variables only stealing money, and output variables only penalty sentence.
As can be seen from beneath the console information, the accuracy is indeed very stagnant, loss card in 0.x level, can be a preliminary judge either the data in question, or else the network generalization ability problem.
Do it again predicted with boston prices data tomorrow.
Meet the last condition.
Impressions today
Thesis accuracy is not high indeed gave me a lot of motivation and confidence, let me re full of energy.
I believe very smoothly after the research will be carried out.