Let Phi(Z_a) be the CDF of the variable x. Find the smallest Z_{a/2}, such that Phi(Z_{a/2}) >= 1 - a/2. mean : N * prob confidence interval : [N * prob - Z_{a/2} * sqrt(N * prob * (1-prob)), N * prob + Z_{a/2} * sqrt(N * prob * (1-prob))]
| 84 | // confidence interval : [N * prob - Z_{a/2} * sqrt(N * prob * (1-prob)), |
| 85 | // N * prob + Z_{a/2} * sqrt(N * prob * (1-prob))] |
| 86 | bool HypothesisTesting(int trails, const vec_pair_t& distribution_to_validate, |
| 87 | const index_map_t& node_count_map) { |
| 88 | int_t count = 0; |
| 89 | float_t prob = 0; |
| 90 | float_t mean = 0; |
| 91 | float_t interval = 0; |
| 92 | int failure_time = 0; |
| 93 | for (const auto& entry : distribution_to_validate) { |
| 94 | auto it = node_count_map.find(entry.first); |
| 95 | if (it == node_count_map.end()) { |
| 96 | count = 0; |
| 97 | } else { |
| 98 | count = it->second; |
| 99 | } |
| 100 | prob = entry.second; |
| 101 | mean = trails * prob; |
| 102 | interval = QUANTILE * std::sqrt(trails * prob * (1 - prob)); |
| 103 | |
| 104 | if (count < mean - interval || count > mean + interval) { |
| 105 | DXINFO("The sampling result of node: %" PRIu64 "(%" PRIu64 |
| 106 | ") is out of the expect range [%f,%f]", |
| 107 | entry.first, count, mean - interval, mean + interval); |
| 108 | failure_time += 1; |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | return failure_time <= MAX_FAILURE_TIME; |
| 113 | } |
| 114 | |
| 115 | } // namespace |
| 116 |