How to estimate parameters for data

Estimation maximum-likelihood
1

Hi, i have used the below code i think i have gone wrong somewhere. can anyone help me to find out where i have gone wrong and how to forward

    @param begin
        tvcl ∈ RealDomain(lower=0, init = 0.6)
        tvv2 ∈ RealDomain(lower=0, init = 40)
        tvQ ∈ RealDomain(lower=0, init = 3)
        #tvk23 ∈ RealDomain(lower=0, init = 3)
        #tvk32 ∈ RealDomain(lower=0, init = 2)
        #tvk20 ∈ RealDomain(lower=0, init = 2.5)
        tvv3 ∈ RealDomain(lower=0, init = 130)
        tvka ∈ RealDomain(lower = 0, init= 0.02)
        Ω ∈ PDiagDomain(init=[0.2])
        σ1 ∈ ConstDomain(1)
        σ2 ∈ RealDomain(lower = 0, init = 0.2)
      end
      @random begin
         η ~ MvNormal(Ω)
      end
      @pre begin
          Ka     = tvka
          CL      = tvcl*exp(η[1])
          Vc      = tvv2
          Q       = tvQ
          #k23     = tvk23
          #k32     = tvk32
          #k20     = tvk20
          Vp      = tvv3
      end
     # @dynamics ImmediateAbsorptionModel
        @dynamics begin
            Depot' =  -Ka*Depot
            Central' =  Ka*Depot - ((Vc/CL)+Q)*Central + Q*(Peripheral)
            #Central' =  Ka*Depot - (k2o+k23)*Central + k32*(Peripheral)
            #Peripheral' = k23*Central - k32*(Peripheral)
            Peripheral' = Q*Central - Q*(Peripheral)
        end
        @derived begin
            depot = Depot
            cp = Central/Vc
            perip = Peripheral
            dv ~ @. Normal(cp, cp*(1+σ2)+σ1)
        end
    end
param = init_param(model)

ylo = CSV.read("D:\\NONMEMnights\\mould\\julia\\yloedited.csv")

data = read_pumas(ylo, dvs=[:dv])

res = fit(model,data,param,Pumas.FOCE())```

```DATA
![image|258x500](upload://ldkBiIK2tIHU24tTfAklhzS03Iw.png) CID	TIME	AMT	DV	MDV
1	0	50	.	1
1	2.5	.	0.41602	0
1	5	.	0.40408	0
1	10	.	0.32019	0
1	15	.	0.36337	0
1	30	.	0.091612	0
1	60	.	0.072907	0
1	90	.	0.056053	0
1	120	.	0.065449	0
1	180	.	0.053612	0
1	240	.	.	1
1	300	.	0.052835	0
1	360	.	0.061585	0
1	480	.	.	1
1	600	.	.	1
2	0	50	.	1
2	2.5	.	0.22745	0
2	5	.	0.48887	0
2	10	.	0.34948	0
2	15	.	0.19076	0
2	30	.	0.17412	0
2	60	.	0.11533	0
2	90	.	0.10667	0
2	120	.	0.088994	0
2	180	.	0.068695	0
2	240	.	.	1
2	300	.	0.065073	0
2	360	.	0.078097	0
2	480	.	0.066583	0
2	600	.	0.051223	0
2

Hi sai_matcha

The code snippet looks incomplete. Could you please link to the complete example and post the error message.

Thanks
Andreas Noack

4 
model = @model begin
    @param begin
        tvcl ∈ RealDomain(lower=0, init = 0.6)
        tvv2 ∈ RealDomain(lower=0, init = 40)
        tvQ ∈ RealDomain(lower=0, init = 3)
        #tvk23 ∈ RealDomain(lower=0, init = 3)
        #tvk32 ∈ RealDomain(lower=0, init = 2)
        #tvk20 ∈ RealDomain(lower=0, init = 2.5)
        tvv3 ∈ RealDomain(lower=0, init = 130)
        tvka ∈ RealDomain(lower = 0, init= 0.02)
        Ω ∈ PDiagDomain(init=[0.2])
        σ1 ∈ ConstDomain(1)
        σ2 ∈ RealDomain(lower = 0, init = 0.2)
      end
      @random begin
         η ~ MvNormal(Ω)
      end
      @pre begin
          Ka     = tvka
          CL      = tvcl*exp(η[1])
          Vc      = tvv2
          Q       = tvQ
          #k23     = tvk23
          #k32     = tvk32
          #k20     = tvk20
          Vp      = tvv3
      end
     # @dynamics ImmediateAbsorptionModel
        @dynamics begin
            Depot' =  -Ka*Depot
            Central' =  Ka*Depot - ((Vc/CL)+Q)*Central + Q*(Peripheral)
            #Central' =  Ka*Depot - (k2o+k23)*Central + k32*(Peripheral)
            #Peripheral' = k23*Central - k32*(Peripheral)
            Peripheral' = Q*Central - Q*(Peripheral)
        end
        @derived begin
            depot = Depot
            cp = Central/Vc
            perip = Peripheral
            dv ~ @. Normal(cp, cp*(1+σ2)+σ1)
        end
    end
param = init_param(model)

ylo = CSV.read("D:\\NONMEMnights\\mould\\julia\\yloedited.csv")

data = read_pumas(ylo, dvs=[:dv])

res = fit(model,data,param,Pumas.FOCE())

this is total code i run. i expected this code will give parameters but it gave same initial estimates value as parameters

5

I edited your text; you need to put the last three back-ticks on a new line :slight_smile:

Can you show us the output of res.optim?

6

result after res.optim

optim
optim1275×272 24.2 KB

7

You are trying to overwrite what’s in res.optim with that command. Could you try to just write res.optim at the command prompt in the window below the script?

8

res.optim gives status:success
optim

9

Please type res.optim in the prompt window instead of the script window.

10

sorry,

res.optim
 * Status: success

 * Candidate solution
    Minimizer: [-5.11e-01, 3.69e+00, 1.10e+00,  ...]
    Minimum:   2.629817e+01

 * Found with
    Algorithm:     BFGS
    Initial Point: [-5.11e-01, 3.69e+00, 1.10e+00,  ...]

 * Convergence measures
    |x - x'|               = 0.00e+00 ≤ 0.0e+00
    |x - x'|/|x'|          = 0.00e+00 ≤ 0.0e+00
    |f(x) - f(x')|         = NaN ≰ 0.0e+00
    |f(x) - f(x')|/|f(x')| = NaN ≰ 0.0e+00
    |g(x)|                 = 1.40e-05 ≤ 1.0e-03

 * Work counters
    Seconds run:   0  (vs limit Inf)
    Iterations:    0
    f(x) calls:    1
    ∇f(x) calls:   1
11

It looks like the likelihood becomes NaN and that is the reason why the optimization stops. I’m not sure why but your error model looks wrong. I assume that you are trying to write an error model with a proportional and additive component. See https://github.com/PumasAI/Pumas.jl/blob/b16c09c6d609ab766a13f79ad65fa711c0125a40/test/nlme/theophylline.jl#L525 for an example of how to do that in Pumas.

1 Like
12

i tried to write both proportional and additive error. ill go through the link, thank you.