> The float epsilon should ideally be greater than the minimum score > increment,
I'm not sure I understand why this is. epsilon is based on 1, (smallest number such that 1-epsilon != 1, right?). What's special about 1? I'm worried about the impact of things like this: smallfloat(10) + smallfloat(1) + smallfloat(1) + smallfloat(1) -> 10 And it makes things very order dependent: smallfloat(1) + smallfloat(1) + smallfloat(1) + smallfloat(10) -> 12 Also, epsilon related to the mantissa, not the exponent? That would make it 1/8, not 1/32. Also, if we don't need to represent very small numbers, we could lower the zero point of the exponent (currently it's 15 for the 5/3 split), right? For reference, here are some of the current byte->float mappings around 1: 116) 0.25 117) 0.3125 118) 0.375 119) 0.4375 120) 0.5 121) 0.625 122) 0.75 123) 0.875 124) 1.0 125) 1.25 126) 1.5 127) 1.75 128) 2.0 129) 2.5 130) 3.0 131) 3.5 132) 4.0 133) 5.0 134) 6.0 135) 7.0 136) 8.0 137) 10.0 138) 12.0 139) 14.0 140) 16.0 141) 20.0 142) 24.0 143) 28.0 144) 32.0 145) 40.0 146) 48.0 147) 56.0 148) 64.0 149) 80.0 150) 96.0 151) 112.0 152) 128.0 -Yonik Now hiring -- http://forms.cnet.com/slink?231706 --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
