function EnvironmentOut=changeToBaseReward(Environment)
%Environment.drugState=(environmentParameters.n_healthy_goals+environmentParameters.n_base_states)+(1:environmentParameters.n_drug_goals);
EnvironmentOut=Environment;
% display('before normalisation')
% displayEnvironment(EnvironmentOut);
% display('press key')
% pause
for st= Environment.drugStates
stpos=st - Environment.drugStates(1)+1;
for action=1:Environment.Num_Actions
v1=EnvironmentOut.reward{st,action}(:);
v2=Environment.baseReward{stpos,action}(:);
% display (EnvironmentOut.nodenames{st})
% display (EnvironmentOut.actionName{action})
% display(v1)
EnvironmentOut.reward{st,action}(:)=Environment.baseReward{stpos,action}(:);
EnvironmentOut.ps{st,action}(:)=Environment.basePs{stpos,action}(:);
% v2=EnvironmentOut.reward{st,action}(:);
% display(v2)
end
end
for st= EnvironmentOut.drugReachabeState
stpos=st - EnvironmentOut.drugReachabeState(1)+1;
for action=EnvironmentOut.toDrugActionIdx
%v3=EnvironmentOut.reward{st,action}(:);
% display (EnvironmentOut.nodenames{st})
% display (EnvironmentOut.actionName{action})
% display(v3)
EnvironmentOut.reward{st,action}=EnvironmentOut.baseRewardBaseState{stpos,action-EnvironmentOut.toDrugActionIdx+1};
% v4=EnvironmentOut.reward{st,action}(:);
% display(v4)
end
end
EnvironmentOut=createInverseTransitions(EnvironmentOut);
% display('back to normal')
% displayEnvironment(EnvironmentOut);
% display('press key')
% pause
end