search

cmu-phil / tetrad

Repository for the Tetrad Project, www.phil.cmu.edu/tetrad.
GNU General Public License v2.0
399 stars 110 forks source link

The use of Tetrad #1795

Open miaomiao-alt opened 1 month ago

miaomiao-alt commented 1 month ago

Hello, let me explain my operation steps in detail: Step 1: Enter the data Step 2: Set the prior knowledge-level constraint (0 level: 47 variables; 1 level: 11 variables) Step 3: Select the algorithm -BOSS (default parameter Settings) Step 4: Input the result graphs and data of the previous three steps into the SEM parameter model - the problem of undirected edges occurs image

cg09 commented 1 month ago

the output is a CPDAG--a collection of DAGs that entail the same conditional independence relations, a Markov Equivalence class. The estimator will only take a DAG. To estimate you need to select a DAG from the output. You can do so by choosing any direction you want for the undirected edges, as long as you do not introduce a new collider---X -> Y<- Z.

On Wed, Jul 10, 2024 at 9:47 PM miaomiao-alt @.***> wrote:

Hello, let me explain my operation steps in detail: Step 1: Enter the data Step 2: Set the prior knowledge-level constraint (0 level: 47 variables; 1 level: 11 variables) Step 3: Select the algorithm -BOSS (default parameter Settings) Step 4: Input the result graphs and data of the previous three steps into the SEM parameter model - the problem of undirected edges occurs image.png (view on web) https://github.com/cmu-phil/tetrad/assets/172348150/add3a77a-8f2e-46e3-8963-e70e5bd4e60c

— Reply to this email directly, view it on GitHub https://github.com/cmu-phil/tetrad/issues/1795, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD4Y3OOCV4DCNORR3SXM2O3ZLXP2JAVCNFSM6AAAAABKV76Y5CVHI2DSMVQWIX3LMV43ASLTON2WKOZSGQYDEMBTHE2TAMI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

miaomiao-alt commented 1 month ago

So how do I achieve this？

cg09 commented 1 month ago

Make a Graph box. Draw the DAG you want in the Graph box. Connect the Graph box to a PM box. Parametrize it in a PM box. Estimate the model in the PM box by connecting it to an Estimate Box. Get the estimate. Watch the tutorial online at the Center for Causal Discovery. Read the manual. Read the Primer by Pearl and Glymour.

On Wed, Jul 10, 2024 at 10:24 PM miaomiao-alt @.***> wrote:

So how do I achieve this？

— Reply to this email directly, view it on GitHub https://github.com/cmu-phil/tetrad/issues/1795#issuecomment-2221864868, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD4Y3OINMPYETLPUC3EMAVDZLXUGZAVCNFSM6AAAAABKV76Y5CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRRHA3DIOBWHA . You are receiving this because you commented.Message ID: @.***>

miaomiao-alt commented 1 month ago

thank you so much！

miaomiao-alt commented 1 month ago

I can understand the advice you gave me: Draw the DAG you want in the Graph box. However, I want to get this DAG graph by setting the prior knowledge and then using the BOSS algorithm. How should I operate this way?

jdramsey commented 1 month ago

Hi @miaomiao-alt, would you like to zoom sometime? I can show you.

miaomiao-alt commented 1 month ago

yes,I can

jdramsey commented 1 month ago

Send me an email.

miaomiao-alt commented 1 month ago

Can you give me a few minutes while I get ready?

miaomiao-alt commented 1 month ago

hemiao2207@163.com

miaomiao-alt commented 1 month ago

I am OK now

jdramsey commented 1 month ago

I sent you an email.

miaomiao-alt commented 1 month ago

How do I know if my model is good or bad？ image

cg09 commented 1 month ago

You don't. All you know is that it was found by a procedure that in the large sample limit converges to the right answer. That is all you know about any statistical estimator.

On Thu, Jul 11, 2024 at 8:34 PM miaomiao-alt @.***> wrote:

How do I know if my model is good or bad？ image.png (view on web) https://github.com/user-attachments/assets/f8eed617-a7c0-4c47-87cc-9e972c23c491

— Reply to this email directly, view it on GitHub https://github.com/cmu-phil/tetrad/issues/1795#issuecomment-2224209504, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD4Y3OPP3DEYERLV3OFF7LLZL4QAXAVCNFSM6AAAAABKV76Y5CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRUGIYDSNJQGQ . You are receiving this because you commented.Message ID: @.***>

jdramsey commented 1 month ago

The stats don't look super-good, but on the other hand, you have included a fair amount of background knowledge. P-value and chi-square are tricky statistics for linear models; if the model isn't essentially perfect; these don't register well. A better tool is the Markov checker that we're developing, which you're free to use, though we don't have a paper for it yet. (Working on it.)

For BIC score, you could try different algorithms or different settings for the algorithms and try to get the best score you you can; this method is often used.

Otherwise, Clark is right; you want to pick an algorithm that has a good performance in general. BOSS is a good choice, or GRaSP.

jdramsey commented 1 month ago

We could zoom again and discuss this a bit. I was fairly busy for the rest of the day yesterday, but later today may have some more time...

jdramsey commented 1 month ago

In my private version of Tetrad I have some tools that could help you choose a good model; I could share this with you if you like. It is an issue I've been thinking about and which is not well-addressed in the current literature.

miaomiao-alt commented 1 month ago

OK! What time are you free today? Let's make an appointment for another meeting. I have a few more questions I'd like to ask you.

jdramsey commented 1 month ago

I'm free at 2 if that will work.

miaomiao-alt commented 1 month ago

Did you say 2:00 p.m., American time? It was two o 'clock in the morning in Beijing. Would you be free in seven or eight hours, that is, at seven or eight o 'clock in the evening US time

---- Replied Message ---- | From | Joseph @.> | | Date | 07/12/2024 23:39 | | To | cmu-phil/tetrad @.> | | Cc | miaomiao-alt @.>, Mention @.> | | Subject | Re: [cmu-phil/tetrad] The use of Tetrad (Issue #1795) |

I'm free at 2 if that will work.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

jdramsey commented 1 month ago

I see, sorry. I could maybe meet at 8 PM EST, would that work?

miaomiao-alt commented 1 month ago

OK！See you in eight hours.

---- Replied Message ---- | From | Joseph @.> | | Date | 07/13/2024 00:14 | | To | cmu-phil/tetrad @.> | | Cc | miaomiao-alt @.>, Mention @.> | | Subject | Re: [cmu-phil/tetrad] The use of Tetrad (Issue #1795) |

I see, sorry. I could maybe meet at 8 PM EST, would that work?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

jdramsey commented 3 weeks ago

@miaomiao-alt Can I close this issue?