Here are some possible datat types may be used.
float32¶
float16 - IEEE 754 half-precision binary floating-point¶
- format: binary16
- sign bit: 1bit
- exponent: 5bits
- significand precision: 11 bits (10 explicitly stored)
* `exponent` is the value of the number that represented by the binaries.
* rule1:
* 0 00000 `1111111111` = (-1)^0 × 2^(-14) x (`1023`/1024) ≈ 0.0000609756
* rule2:
* 0 01111 `0000000001` = (-1)^0 × 2^(15-15) × (1 + `1`/1024) = 1.0009765625
* 0 01101 `0101010101` = (-1)^0 × 2^(13-15) × (1 + `341`/1024) = 0.333251953125 ≈ 1/3
* rule3:
* 0 11111 0000000000 = infinity