string syntax for umxRAM about umx

[bwiernik]

I agree that either graphviz or sem has the most intuitive syntax. sem exactly would be convenient to make it maximally easy to re-use scripts in umx. A reasonable compromise would be to throw a warning if a string syntax model is fit without latent/manifest variances, etc.

Functions for umx_lavaan2RAM, umx_lavaan2RAM, and umx_lavaan2RAM would be really nice for adapting published scripts. These could perhaps return lists of umx paths. They should include the automatic paths assumed by the various programs, with warnings that they were added. Positioning these functions as conversion tools, not recommended for building one’s own models, would be good.

[tbates]

Decided to treat the lavaan syntax as a universal grammar for SEM, and add ability to process this in umx via umxRAM

[tbates]

One option is to create umx_lavaan2RAM() and let people use that. Trouble is lavaan, like mplus, has defaults, so that code would involve more than just translating what the user requested, but also what lavaan decided their request implied (like variances for manifests etc. I've always avoided that it only seems smart for about a day...

I am thinking now that the syntax I most like playing with is graphviz, paradoxically also invented at AT&T by the guys who did R.

It's very sem like, so just:

    A -> B

Which obviates typing all the functions, parameter names, commas and = signs and closing brackets and trailing commas that frustrate people

A nice thing might be to allow abbreviation lists, so that for instance instead of typing:

    umxRAM(paths = "
        Stress <-> Extraversion,
        Extraversion -> Depression
        Stress -> Depression"
    )

One could say

    umxRAM(vars = c(S = Stress, E = "Extraversion", D = "Depression), paths = "
        S <-> E,
        S E -> D"
    )

Both are probably easier than:

    umxRAM("myModel",
        umxPath("Stress", with = "Extraversion"),
        umxPath(c("Stress", "Extraversion"), to = "Depression")
    )

Which in turn beats

    mxModel("myModel", type = "RAM",
       manifests = c("Stress", "Extraversion", "Depression") 
        mxPath("Stress", to = "Extraversion", arrows= 2),
        mxPath(c("Stress", "Extraversion"), to = "Depression")
    )

[mcneale]

Good plan! Is it best to go Lavaan -> Umx or directly Lavaan -> OpenMx?

[tbates]

The lavaan is processed using a mix of umx or OpenMx calls as is easiest.

It all just comes back as an MxRAMModel with sub models if it's multi-group

But a lot of umx to get the auto-means, auto data-set trimming, labels, start values, group= facility etc etc.