#include <bits/stdc++.h>
//#include <chrono>
//#pragma GCC optimize("O3")
using namespace std;
#define reps(i,s,n) for(int i = s; i < n; i++)
#define rep(i,n) reps(i,0,n)
#define Rreps(i,n,e) for(int i = n - 1; i >= e; --i)
#define Rrep(i,n) Rreps(i,n,0)
#define ALL(a) a.begin(), a.end()

using ll = long long;
using vec = vector<ll>;
using mat = vector<vec>;

ll N,M,H,W,Q,K,A,B;
string S;
using P = pair<ll, ll>;
const ll INF = (1LL<<60);

template<class T> bool chmin(T &a, const T &b){
    if(a > b) {a = b; return true;}
    else return false;
}
template<class T> bool chmax(T &a, const T &b){
    if(a < b) {a = b; return true;}
    else return false;
}

template <unsigned long long mod > class modint{
public:
    ll x;
    constexpr modint(){x = 0;}
    constexpr modint(ll _x) : x((_x < 0 ? ((_x += (LLONG_MAX / mod) * mod) < 0 ? _x + (LLONG_MAX / mod) * mod : _x) : _x)%mod){}
    constexpr modint set_raw(ll _x){
        //_x in [0, mod)
        x = _x;
        return *this;
    }
    constexpr modint operator-(){
        return x == 0 ? 0 : mod - x;
    }
    constexpr modint& operator+=(const modint& a){
        if((x += a.x) >= mod) x -= mod;
        return *this;
    }
    constexpr modint operator+(const modint& a) const{
        return modint(*this) += a;
    }
    constexpr modint& operator-=(const modint& a){
        if((x -= a.x) < 0) x += mod;
        return *this;
    }
    constexpr modint operator-(const modint& a) const{
        return modint(*this) -= a;
    }
    constexpr modint& operator*=(const modint& a){
        (x *= a.x)%=mod;
        return *this;
    }
    constexpr modint operator*(const modint& a) const{
        return modint(*this) *= a;
    }
    constexpr modint pow(unsigned long long pw) const{
        modint res(1), comp(*this);
        while(pw){
            if(pw&1) res *= comp;
            comp *= comp;
            pw >>= 1;
        }
        return res;
    }
    //以下、modが素数のときのみ
    constexpr modint inv() const{
        if(x == 2) return (mod + 1) >> 1;
        return modint(*this).pow(mod - 2);
    }
    constexpr modint& operator/=(const modint &a){
        (x *= a.inv().x)%=mod;
        return *this;
    }
    constexpr modint operator/(const modint &a) const{
        return modint(*this) /= a;
    }
};
#define mod1 998244353
using mint = modint<mod1>;

ostream& operator<<(ostream& os, const mint& a){
    os << a.x;
    return os;
}
using vm = vector<mint>;

class NTT{
    static int root;
    static vm root_pow;
    static vector<int> id;

    static void make_root_pow(int n){
        if(n + 1 == (int)root_pow.size()) return;
        root_pow.resize(n + 1);
        mint new_root = mint(root).pow((mod1 - 1) / n);
        root_pow[0].x = 1;
        rep(i,n){
            root_pow[i + 1] = root_pow[i] * new_root;
        }
    }
    static void make_bit_reverse(int n){
        if(n == (int)id.size()) return;
        id.resize(n);
        iota(ALL(id), 0);
        for(int i = 1; (1<<i) <= n; ++i){
            int l = 1<<(i - 1), r = 1<<i;
            int plus = n >> i;
            for(int j = l; j < r; ++j){
                int temp = id[j - l] + plus;
                if(j < temp) swap(id[j], id[temp]);
            }
        }
    }
    static void dft(int n, vm &f, bool inv){
        vm g(n);
        rep(i,n) g[i] = f[id[i]];
        swap(f, g);
        for(int l = n / 2, len = 1; l >= 1; l /= 2, len *= 2){
            for(int i = 0; i < n; i += len * 2){
                rep(j, len){
                    mint z_f = (inv ? root_pow[n - l * j] : root_pow[l * j]) * f[i + len + j];
                    g[i + j] = f[i + j] + z_f;
                    g[i + len + j] = f[i + j] - z_f;
                }
            }
            swap(f, g);
        }
        if(inv) {
            mint n_inv = mint(n).inv();
            rep(i, n) f[i] *= n_inv;
        }
    }
public:
    NTT(int _x = 3) {
        root = _x;
    }
    void dft_2D(int n, int m, vector<vm> &a, bool inv){
        //簡単に、書き換える形で
        //aがn×mサイズであることや、n,mが2冪であることは仮定
        make_root_pow(m);
        make_bit_reverse(m);
        rep(i, n) dft(m, a[i], inv);
        make_root_pow(n);
        make_bit_reverse(n);
        rep(j, m){
            vm temp(n);
            rep(i, n) temp[i] = a[i][j];
            dft(n, temp, inv);
            rep(i, n) a[i][j] = temp[i];
        }
    }
    static vm convolution(vm &a, vm &b, int size_a = INT_MAX, int size_b = INT_MAX){
        if(size_a > (int)a.size()) size_a = (int)a.size();
        if(size_b > (int)b.size()) size_b = (int)b.size();
        int sz = size_a + size_b - 1, n = 1;
        while(sz > n) n *= 2;
        vm g(n), h(n), gh(n);
        copy(a.begin(), a.begin() + size_a, g.begin());
        copy(b.begin(), b.begin() + size_b, h.begin());
        make_root_pow(n);
        make_bit_reverse(n);
        dft(n, g, false);
        dft(n, h, false);
        rep(i, n) gh[i] = g[i] * h[i];
        dft(n, gh, true);
        //gh.resize(sz);
        gh.resize(size_a);
        return gh;
    }
};
int NTT::root;
vm NTT::root_pow;
vector<int> NTT::id;

const ll MAX_N = ll(4e+5) + 10;
vm fact(MAX_N, mint(1)), fact_inv(MAX_N, mint(1)), n_inv(MAX_N, mint(1));
void makefact(){
    mint tmp;
    reps(i,2,MAX_N) fact[i] = fact[i-1] * tmp.set_raw(i);
    fact_inv[MAX_N - 1] = fact[MAX_N - 1].inv();
    Rreps(i, MAX_N - 1, 1){
        fact_inv[i] = fact_inv[i + 1] * tmp.set_raw(i + 1);
        n_inv[i + 1] = fact[i] * fact_inv[i + 1];
    }
}
mint nCm(ll n, ll m){
    return fact[n] * fact_inv[n-m] * fact_inv[m];
}
mint nCm_inv(ll n, ll m){
    return fact[n-m] * fact[m] * fact_inv[n];
}

int main(){
    makefact();
    cin>>N;
    vm pol(N - 1), res_v(N - 1, 0);
    rep(i, N - 1) pol[i] = fact_inv[i] * (i + 1);
    res_v[0] = 1;
    int n = N;
    NTT ntt;
    while(n){
        if(n&1) res_v = ntt.convolution(res_v, pol);
        pol = ntt.convolution(pol, pol);
        n >>= 1;
    }
    cout<<fact[N - 2] * mint(N).pow(N - 2).inv() * res_v[N - 2]<<endl;
}