<Although the "window" will slide, I don't believe that the net has assimilated the new data at that point.>
If you have been sitting in front of my computer yesterday you have have drawn a different conclusion. I am not saying the net has assimilated the data completely, all I'm saying is the net that is producing the predictions in the verify period is not the same net that was trained. It contains data that was not trained.
< I retrain my nets on the weekends but use the batch files to make predictions during the week to avoid the situation you're talking about.>
Even though you don't see it happen, the program opens the net, reads the data files, and then reads the result for the report. The net still sees the new data just the same as if you would have done it manually.
<Jay, it seems from his response that by not training the net after updating the files will solve your delemna. Just hit the predict button.>
Like I've already said, from what I have seen happen lately, I am beginning to even question Andrew's comments. I am starting to curse the addition of this new computer. All the tests that I have done in the last few days would have taken weeks on the 166, and therefore I would have never done them.
I believe I have found a better way of maintaining the integrity of the original net. Before starting to train a new net, remove the latest line of data from the target stock's data file. Set up the net the way you want it. Then add the line of data back into the file and have Neuro reread the data files. The training start and end dates will move ahead one day. Train the net completely. Before updating the data files open the net and move the training start and end dates back one day. Update the files. When Neuro rereads the updated data, the training dates will move to there original location and the original net will be restored.
I know, I'm losing it.
Jay |
