#pragma GCC optimize("Ofast")
#pragma GCC optimize("unroll-loops")
#include<bits/stdc++.h>
using namespace std;
typedef long long ll;
typedef long double ld;
typedef pair<ld,ld> pd;
typedef pair<ll,ll> pl;
typedef pair<ll,pl> ppl;
const ll lim=3e5+5,mod=1e9+7;
ll par[lim],a[lim],dp[lim]; //Calced vals can be updated later. In a strict DP sense it would be dp(pos, time) rather than just dp(pos)
ll fs(ll x){
if(x==par[x])return x;
else return par[x]=fs(par[x]);
}
void mergeset(ll a,ll b){
a=fs(a),b=fs(b);
if(a==b)return;
par[a]=b;
}
int main(){
ios_base::sync_with_stdio(0);cin.tie(0);
ll n,d;
cin>>n>>d;
vector<pl>v;
for(int i=0;i<n;i++){
cin>>a[i];
par[i]=i;
v.emplace_back(a[i],-i);
}
sort(v.begin(),v.end());
for(auto p:v){
ll i=-p.second;
dp[i]=max(1ll,dp[i]);
for(int j=max(i-d,0ll);j<i;j++)dp[i]=max(dp[i],dp[j]+1);
ll t = fs(i); //t is the rightmost node to which we can push a dp-val update
while(t < min(i+d,n-1)){
mergeset(t, t+1);
t = fs(t);
}
for(int j=i+1;j<=t;j++){
if(dp[j])dp[j]=max(dp[j],dp[i]);
}
//cout<<"after "<<i<<" (val "<<p.first<<"):\n";
//for(int i=0;i<n;i++)cout<<dp[i]<<" ";
//cout<<"\n";
}
cout<<dp[n-1];
return 0;
}
