Lula 3d Patch Fr
August 2018, 2565. 
A:
You have a very tricky line of work. 
Converting a multi-part number (like 1,000,500) into a single, non-decimal string (like "1,000,500")
Converting that into an exact, but only approximate, decimal equivalent (e.g. "1,000.5" in this case)
Providing this output to a very strict, non-imposing, audience: a machine learning algorithm that doesn't understand that you've done any manipulation.
So, by doing some of these things, you open yourself up to accidentally introducing errors into your conversion process. If you don't do the conversion correctly, you may end up with errors that would lead to your model incorrectly predicting the wrong output.
As an example of this error, you mentioned "1,000.5" in your question. If your data contains multiple values like this, you may end up with one of the values ending up with a decimal point that is not in your format string. That's a problem, and that may be the source of your issues.
Even if you do the conversion correctly, the math of probability is not straightforward. Your probability of a correct outcome, i.e. the probability of having the machine learning model get the correct answer, depends on the accuracy of your conversion function. If the output of the conversion function is close to the correct answer, but not exactly the correct answer, you are introducing a bias into the probability of the model getting the right answer.
So, to convert those numbers into decimal approximations, you need to convert them to percentages, which is what you did in the "simple conversion" function. That's fine, but be aware that the conversion is not 100% accurate. The reason why you don't get 100% accuracy is that even though you have "1,000,500" converted into "1,000.5", it's actually rounded up to 1,000.5. It doesn't get to the exact accuracy of "1,000.5", but it is close enough.
That introduces another error: rounding. If you round the 1,000.5 up to 1,000.6, you are converting it from "closer to the correct answer than what is currently given" to "closer to the correct answer than what is currently given". That's a really subtle error, and it's also a
44926395d7

Related