Get optimal point and estimated values at arbitrary points

[ilario]

Thanks for the very nice software!

We are using it for the optimization of the synthesis of a catalyst (the objective is the activity of the catalyst).

I was wondering if two things are possible:

• get the estimated optimum position from a model. I managed to get the software to suggest new experiments, but I cannot get the estimated location of the best point.
• probe the model at arbitrary points (for plotting the model as heatmaps, for example, see the plots in Figure 8 here: https://pubs.acs.org/doi/10.1021/acsnano.8b04726 ).

In order to have that, I suppose that there could be a function returning the model, so that the user can interact with it directly and fetch this information.

[marcosfelt]

Hi Ilario! I would love those two features as well, so I'd be willing to work on getting them written. Which strategy are you using? It would be easiest to start with the strategies based on BOtorch such as TSEMO.

…

On Fri, 14 Oct 2022 at 15:33, Ilario Gelmetti ***@***.***> wrote: Thanks for the very nice software! We are using it for the optimization of the synthesis of a catalyst (the objective is the activity of the catalyst). I was wondering if two things are possible: - get the estimated optimum position from a model. I managed to get the software to suggest new experiments, but I cannot get the estimated location of the best point. - probe the model at arbitrary points (for plotting the model as heatmaps, for example, see the plots in Figure 8 here: https://pubs.acs.org/doi/10.1021/acsnano.8b04726 ). In order to have that, I suppose that there could be a function returning the model, so that the user can interact with it directly and fetch this information. — Reply to this email directly, view it on GitHub <#214>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGF3ORZWOJOIHKK36BIXYTDWDFVKRANCNFSM6AAAAAARFJQRW4> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ilario]

Hi!
For now we just tried all of the strategies, without applying them in our experimental workflow yet.
I didn't get to use TSEMO due to #213.

Also MTBO seems based on BOtorch but I am not sure it makes sense for our usecase: seems that it requires some experimental data from a different experimental setup (MIT_case1 from the example) and some new data with the interesting one (MIT_case2 from the example) but we just have collected preliminary data for one experimental setup (our experimental prev_res). Did I understand well?

[ilario]

Another question: is it fair to assume that the best point, according the used model, is the first point in the suggest_next_locations list?

[marcosfelt]

On MTBO, you can use it in the case of having only experimental data from another case. There, MTBO will suggest a random experiment at first and then improve. I'll be merging a new version of MTBO in soon based on some of our recent internal work. The best point is not what is suggested always. So, you should use `exp.data` to find that.

…

On Mon, 17 Oct 2022 at 15:30, Ilario Gelmetti ***@***.***> wrote: Another question: is it fair to assume that the best point, according the used model, is the first point in the suggest_next_locations list? — Reply to this email directly, view it on GitHub <#214 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGF3ORZXPZGVR2W2VCHNDHLWDVPIJANCNFSM6AAAAAARFJQRW4> . You are receiving this because you commented.Message ID: ***@***.***>