How to make a precalculated lookup table for normal CDF ?

[Naucle]

Hello,

I have a program that runs the normal CDF function up to 240 000 000 times, which makes it a little bit slow, and I have been adivised to precalculat a lookup table and to interpolate from it, but I'm just a novice in C++ and I have no idea how to do that (google couldn't help me neither). Do you people have any useful links ?

Thank you.

 

[mhy]

You can just use a straight c++ map.

(Pseudocode)
map<float, float> n_map;

min = -10;
max = 10;
granularity = 0.05;

for (int x = min ; x < max; x += granularity) {
n_map[x] = normal_cdf(x);
}

If you want to avoid using a map & memory for it you will probably need something cuter

 

[ExSan]

Hello,

I have a program that runs the normal CDF function up to 240 000 000 times, which makes it a little bit slow, and I have been adivised to precalculat a lookup table and to interpolate from it, but I'm just a novice in C++ and I have no idea how to do that (google couldn't help me neither). Do you people have any useful links ?
Thank you.

I do not know what is
a precalculated lookup table for normal CDF
could you please point a link that would describe what you are talking about.
thanks !

 

[Naucle]

You can just use a straight c++ map.

(Pseudocode)
map<float, float> n_map;

min = -10;
max = 10;
granularity = 0.05;

for (int x = min ; x < max; x += granularity) {
n_map[x] = normal_cdf(x);
}

If you want to avoid using a map & memory for it you will probably need something cuter

That's what I did actually, but it's still tricky to interpolate and the precision is not that good.

 

[mhy]

Well, here's a few other options:

1) Find a really fast implementation of the CDF function.
2) Create your own CDF function with memoization / approximation, etc (so if tolerance is within 0.001 for example, do not recalculate)

You will need some notion of tolerance either way unless you plan on recalculating every time...