Use of Naive Bayes to predict price direction in R

ironchef said:
Thank you for sharing. May I ask you a couple of questions:

1. This:
is just a set of "rules"?

2. This:
is the learning algorithm? It is just Bayesian statistics. So you apply a probability to the parameters you use? This is no different from say using Kalman filter algorithm, I don't see any "learning" involved.

3. Any bias/drift of the data set will produce a non 50/50 outcome?

4. In my simplistic view, learning means I change the parameters when I find them not useful, not by adding noise term to the parameters?

Not sure if I make any sense, if not, just ignore my questions.

Thank you.
More...

Answering your questions.
1- Is just a set of rules?
Not really. In machine learning, you go from independent variables to dependent variables. Meaning your machine learning model tries to predict the dependent variable by using a set of independent variables. So those were not rules, but they were the independent variables/features that I chose.

2- Yes, it is a machine learning algorithm. It is not Bayesian statistics per say. Naibe Bayes makes use of conditional probability with the assumption of independence to predict your classes (from your independent variable). It is mostly use for classification problems. In my case, predicting whether next day's candle will be a bullish or bearish candle.

4- Take the following example:
say if the hypothesis is that if the price closes above ema7 and the previous'day candle was bullish, the odds are high the next day candle will be bullish.
I could use a learning model to see whether such representation exists in my data. If the model performs poorly, I might use the ema20 and maybe add MACD all these as new independent variables. I might even modify if I want. But all these with the aim of finding a model that has learned well enough and is able to perform well on out of sample data.
 
arsene007 said:
Answering your questions.
1- Is just a set of rules?
Not really. In machine learning, you go from independent variables to dependent variables. Meaning your machine learning model tries to predict the dependent variable by using a set of independent variables. So those were not rules, but they were the independent variables/features that I chose.

2- Yes, it is a machine learning algorithm. It is not Bayesian statistics per say. Naibe Bayes makes use of conditional probability with the assumption of independence to predict your classes (from your independent variable). It is mostly use for classification problems. In my case, predicting whether next day's candle will be a bullish or bearish candle.

4- Take the following example:
say if the hypothesis is that if the price closes above ema7 and the previous'day candle was bullish, the odds are high the next day candle will be bullish.
I could use a learning model to see whether such representation exists in my data. If the model performs poorly, I might use the ema20 and maybe add MACD all these as new independent variables. I might even modify if I want. But all these with the aim of finding a model that has learned well enough and is able to perform well on out of sample data.
More...
Thank you. I am very interested in AI but always wonder what that meant. You gave me something tangible to work on. Appreciate it.
 
You must log in or register to reply here.
Back
Top