Topic List | Page List: 1 |
---|---|
Topic | My AI predicts these are the future prices of bitcoin |
Yellow 08/31/19 5:32:18 AM #22: | I've found that long-term memory NNs don't take into account long-term "inflation" very well. Aka, a currency that was $5 a year ago, worth $100 today, completely throws off everything the NN knows about the currency, causing it to predict a random value between $5 and $100. This is because it's predicting absolute values, with no sense of relativity. In our minds, the pattern may carry over, but in this NN 200 is a different dimension than 20. The goal here is not to predict the exact values, it's to predict the value relative to the previous value. The solution might be to generate a dataset that has the values in differences by percentage points. instead of 100, 100, 200, 150 it could be n/a, 100%, 200%, 75% This would eliminate the factor of inflation. It makes me think that a any proper NN should always be used to predict relative values, never absolute ones. That makes more sense, a real human always thinks in relative terms. Sardanapallus posted... Yellow posted...If humans can do it, a neural network can do it. And if it is possible the world will possibly look very different in the future. --- ... Copied to Clipboard! |
Topic List | Page List: 1 |