14:31:27	 From Yogesh D : We are starting.
15:19:54	 From Siva Athreya : question is not entirely clear
15:20:47	 From MK : he means a joint distribution with given marginals. Copulas are the answer
15:33:33	 From Utpal Chattopadhyay : Does your ABM say anything about Dharavi type situations?
15:47:42	 From thattai : 1. One issue is the stochastic variation in the time to these two options. E.g. consider a scenario where most deaths occur early, but a few occur after prolonged treatment after D days. Depending on the way “recovery” is defined (if it is not by a negative Covid test, which many states don’t require) then you can only be designated as recovered after D days, since before that time there is a non-zero death probability. 2. If the average days to recovery is different from the average days to death (conditioned on the final outcome) then the measured counts are reporting on original infections that initiated at different times. Since we are not in steady state, this biases the outcome. 3. In exponential steady state, after transients, the ratio of deaths/day to recoveries/day is a meaningful number. 4. If we had detailed dates of infection and final outcome (death/recovery) for each individual, then we can of course define a hazard function in the usual way, which is the right way to go about it.
16:32:39	 From Siva Athreya : How is T determined ?
16:34:54	 From Rajesh Sundaresan, IISc : This looks like an G/G/infinity discrete-time queue. Arrivals are L. Departures are M (in a slot). Service times (middle G) are given the p1, p2, … p_T. Perhaps ML estimation for the arrivals? 
16:36:04	 From aditya : @rajesh in fact we've heard that testing (in labs) actually follows a queueing model -- each lab implements a (priority queue). so this may be relevant indeed
16:37:27	 From Navin : the p distribution is the incubation period distribution. this has been studied for covid.
16:37:44	 From Srikanth Iyer : Instead of minimising the oscillation can we force the L_i to follow some model such as the SIR?
16:39:33	 From Navin : @srikanthiyer you can certainly do many other things. but even naive things seem to work quite well in experiments.
16:55:42	 From D Manjunath : Are the L simulated to be independent?
16:59:24	 From MK : You can input L to be any numbers - no restriction. In the simulation he used simulation to generate L_is