#include <bits/stdc++.h>

using namespace std;
typedef long long ll;
typedef pair<ll, ll> p_ll;

template<class T>
void debug(T itr1, T itr2) { auto now = itr1; while(now<itr2) { cout << *now << " "; now++; } cout << endl; }
#define repr(i,from,to) for (ll i=(ll)from; i<(ll)to; i++)
#define all(vec) vec.begin(), vec.end()
#define rep(i,N) repr(i,0,N)
#define per(i,N) for (int i=(int)N-1; i>=0; i--)

const ll MOD = 998244353;
const ll LLINF = pow(2,61)-1;
const int INF = pow(2,30)-1;

vector<ll> fac;
void c_fac(int x=pow(10,6)+10) { fac.resize(x,true); rep(i,x) fac[i] = i ? (fac[i-1]*i)%MOD : 1; }
ll modpow(ll x, ll p) { ll result = 1, now = 1, pm = x; while (now<=p) { if (p&now) { result = result * pm % MOD; } now*=2; pm = pm*pm % MOD; } return result; }
ll inv(ll a, ll m=MOD) { ll b = m, x = 1, y = 0; while (b!=0) { int d = a/b; a -= b*d; swap(a,b); x -= y*d; swap(x,y); } return (x+m)%m; }
ll nck(ll n, ll k) { return fac[n]*inv(fac[k]*fac[n-k]%MOD)%MOD; }
ll gcd(ll a, ll b) { if (a<b) swap(a,b); return b==0 ? a : gcd(b, a%b); }
ll lcm(ll a, ll b) { return a/gcd(a,b)*b; }

// ----------------------------------------------------------------------
// ----------------------------------------------------------------------

struct FastNumberTheoreticTransform {

  ll size, size_v12, root, vsize, base;;
  vector<ll> theta;

  FastNumberTheoreticTransform(ll N=20) { 
    size = 1<<N; root = modpow(3,MOD/size);
    theta.resize(size);
    rep(i,size) theta[i] = i==0 ? 1 : theta[i-1]*root % MOD;
  }

  void set(vector<ll> &v1, vector<ll> &v2) {
    size_v12 = v1.size() + v2.size() - 1;
    vsize = 0; while ((1<<vsize)<size_v12) vsize++;
    base = size>>vsize;
  }

  vector<ll> convolution(vector<ll> &x, bool rev, ll depth=0, ll pos=0) {
    if (depth==vsize) return {pos<x.size()?x[pos]:0};
    vector<ll> ex = convolution(x,rev,depth+1,pos);
    vector<ll> ox = convolution(x,rev,depth+1,pos+(1<<depth));
    ll N = 1<<(vsize-depth);
    vector<ll> result(N);
    rep(i,N) result[i] = (ex[i%(N/2)] + theta[base*(1<<depth)*i] * ox[i%(N/2)]) % MOD;
    if (rev && depth==0) {
      reverse(result.begin()+1, result.end());
      ll invs = inv(1<<vsize);
      for (auto &x: result) x = x * invs % MOD;
    }
    return result;
  };

  vector<ll> mul(ll N, vector<ll> &v1, vector<ll> &v2) {
    set(v1, v2);
    vector<ll> cv1 = convolution(v1, false), cv2 = convolution(v2, false);
    vector<ll> cv12(1<<vsize); rep(i,1<<vsize) cv12[i] = cv1[i] * cv2[i] % MOD;
    vector<ll> v12 = convolution(cv12, true);
    v12.resize(N);
    return v12;
  }

};

// ----------------------------------------------------------------------
// ----------------------------------------------------------------------


int main() {
  ll N; cin >> N;

  c_fac();
  vector<ll> kei(N); rep(i,N) kei[i] = (i+1)*inv(fac[i])%MOD;

  // debug(all(kei));
  FastNumberTheoreticTransform fntt;
  vector<ll> result(N,0); result[0] = 1;
  ll np = 1;
  while (np<=N) {
    if (np&N) result = fntt.mul(N,result,kei);
    kei = fntt.mul(N,kei,kei);
    np<<=1;
  }
  ll ho = fac[N-2] * inv(modpow(N,N-2)) % MOD;
  rep(i,N) result[i] = result[i] * ho % MOD;
  // debug(all(result));
  cout << result[N-2] << endl;
  return 0;
}