Possible loss of precision during conversion of float number

Question

I have google on how to get 2 decimal for a float number in java. Below are my codes. Finally here float totalWeight = 0.1*levinWeight+0.8*lsmWeight; I get the error of possible loss of precision ? I would want to first covert the string into float and then have it to be 2 decimal places.

float levinWeight = Float.parseFloat(dataOnlyCombine[2]);
float lsmWeight = Float.parseFloat(dataOnlyCombine[3]);
DecimalFormat df = new DecimalFormat("#.##");
levinWeight = Float.valueOf(df.format(levinWeight));
lsmWeight = Float.valueOf(df.format(lsmWeight));
float totalWeight = 0.1*levinWeight+0.8*lsmWeight;

Answer 1

If you are concerned about precision

• don't use float, it has the lowest precision of any option available. I suggest using double or BigDecimal
• use operation which involve values which can be accurately represented. 0.1 * x will give you error because 0.1 cannot be represented precisely. Using x / 10.0 will have less error.

I would write something like this

double levinWeight = Double.parseDouble(dataOnlyCombine[2]);
double lsmWeight = Double.parseDouble(dataOnlyCombine[3]);
double totalWeight = (levinWeight + 8 * lsmWeight) / 10.0;
// perform rounding only at the end as appropriate.

// to round to two decimal places 
double totalWeight2 = Math.round(totalWeight * 100) / 100.0;

Answer 2

float levinWeight = Float.parseFloat(dataOnlyCombine[2]);
float lsmWeight = Float.parseFloat(dataOnlyCombine[3]);
float totalWeight = 0.1*levinWeight+0.8*lsmWeight;
DecimalFormat df = new DecimalFormat("#.##");
String totalWeightValue = df.format(totalWeight);

Answer 3

If you really want to do it like that, then use BigDecimal. Those floating point classes are perfect for precision. Take a look at them:

• http://voidexception.weebly.com/java-bigdecimal---dealing-with-high-precision-calculations.html
• http://docs.oracle.com/javase/7/docs/api/java/math/BigDecimal.html
• http://www.javaworld.com/article/2075315/core-java/make-cents-with-bigdecimal.html

Default IEEE 746 floating points will not suit your needs. Alternatively, you could use integers and thread them factor 100. So:

• 100 is equivalent to 1.00
• 452 is equivalent to 4.52
• 1 is equivalent to 0.01