How to ensure sum of "sys_probs" equals "1"

[Epikurz]

To generate input files , I have wrote a script in which folders of datasets prepared by dpdata.py and the sampling probability were inputs. With this script, I assign probability of training to the folders, while the probabilities of each sub-folder determined by the number of frames.
Due to the computing mechanism of floats in python, sometimes the sum of sys_probs generate by my scripts did not equal "1" and would raise error in deepmd-kit. Is there any methods to solve the problem?
My script:

import sys
import os
import json

rawfolders = sys.argv[1:]

ratio_prob = input("relative of probability of the data_sets:")
ratio_prob = ratio_prob.strip().split()
ratio_list = [float(x) for x in ratio_prob[:len(sys.argv[1:])]]
relative_prob = [x/sum(ratio_list) for x in ratio_list]
print(relative_prob)
datasets = []
prob = []

for index, ff in enumerate(rawfolders):
presys = os.listdir(ff)
nfrs = []
for i in presys:
targets = os.path.join(ff,i)
if os.path.isdir(targets) and os.path.isfile("%s/force.raw" %(targets)):
datasets.append(targets)
tmp = open("%s/force.raw" %(targets),"r").readlines()
nfr = len(tmp)
nfrs.append(nfr)
temp_prob = [ relative_prob[index]*x/sum(nfrs) for x in nfrs]
prob = prob + temp_prob
prob = [round(x/sum(prob),4) for x in prob]
print(sum(prob))
prob[prob.index(max(prob))] += 1.0 - sum(prob)
print(sum(prob))
alldata = dict(zip(datasets,prob))
print(alldata)

with open("/home/user/input/deepmd/train_input.json","r") as tp:
input = json.load(tp)
input["training"]["training_data"]["systems"] = datasets
input["training"]["training_data"]["sys_probs"] = prob
input["training"]["validation_data"]["systems"] = datasets
input["training"]["validation_data"]["sys_probs"] = prob
with open("newinput.json",'w') as new:
json.dump(input,new,indent=4)

[njzjz]

You can keep a certain number of digits. Also, maybe we should use np.isclose(np.sum(ret_prob), 1) instead of np.sum(ret_prob) == 1. May I ask how much the float error is?

[Epikurz]

Thanks for response！
In most cases，I use round(x,5) and the sum of sys_probs is 1.0004