#include<bits/stdc++.h>
using namespace std;
#define all(fl) fl.begin(),fl.end()
#define pb push_back
#define fi first
#define se second
#define for1(i,j,k) for(int i=j;i<=k;i++)
#define for2(i,j,k) for(int i=j;i>=k;i--)
#define for3(i,j,k,l) for(int i=j;i<=k;i+=l)
#define lb lower_bound
#define ub upper_bound
#define sz(a) (int)a.size()
#define pii pair<int,int>
#define pli pair<long long,int>
#define gcd __gcd
#define lcm(x,y) x*y/__gcd(x,y)
#define pil pair<int,long long>
#define pll pair<long long,long long>
#define eb emplace_back
#define int long long
const long long inf=1e15;
const int maxn=1e5+9;
long long f[1<<15];
int a[maxn][15],b[maxn][15];
int cnt[15];
vector<int>id[15];
string s;
vector<int>bit[1<<15];
long long g(int gr,int k,int mask){
long long cost=0;
int mp=sz(id[gr]);
for (auto v:bit[mask]){
if (mp==0)continue;
if (k==0){
cost+=2*b[id[gr][0]][v];
}
else if (k==mp){
cost+=2*a[id[gr].back()][v];
}
else {
int id1=id[gr][k-1],id2=id[gr][k];
cost+=2*a[id1][v]+2*b[id2][v];
}
}
cost+=k*(k-1)/2;
cost+=(mp-k)*(mp-k-1)/2;
return cost;
}
signed main(){
ios_base::sync_with_stdio(0);
cin.tie(0);
//freopen("temp.INP","r",stdin);
//freopen("temp.ANS","w",stdout);
f[0]=0;
cin>>s;
int n=sz(s);
s=" "+s;
for1(i,1,n){
for1(j,0,14){
a[i][j]=cnt[j];
}
cnt[s[i]-'A']++;
id[s[i]-'A'].pb(i);
}
memset(cnt,0,sizeof(cnt));
for2(i,n,1){
for1(j,0,14){
b[i][j]=cnt[j];
}
cnt[s[i]-'A']++;
}
for1(i,0,14){
for1(j,1,sz(id[i])-1){
for1(k,0,14){
a[id[i][j]][k]+=a[id[i][j-1]][k];
}
}
for2(j,sz(id[i])-2,0){
for1(k,0,14){
b[id[i][j]][k]+=b[id[i][j+1]][k];
}
}
}
for1(i,0,(1<<15)-1){
for1(j,0,14){
if (i>>j&1)bit[i].pb(j);
}
f[i]=inf;
}
f[0]=0;
for1(i,0,(1<<15)-1){
if (f[i]>=inf)continue;
for1(j,0,14){
if (i>>j&1)continue;
int l=0,r=sz(id[j]);
while (l!=r){
int mid=(l+r)/2;
if (g(j,mid,i)>g(j,mid+1,i)){
l=mid+1;
}
else r=mid;
}
f[i+(1<<j)]=min(f[i+(1<<j)],f[i]+g(j,l,i));
}
}
long double answ=f[(1<<15)-1];
answ=answ*0.5;
cout<<setprecision(4)<<fixed<<answ;
}
