hlzhang109 / transtee Goto Github PK
View Code? Open in Web Editor NEWLicense: MIT License
License: MIT License
Based on your detailed code files, I try to reproduce the results about multiple treatmentments and dosages (TCGA) by directly running the commands you provided. I have some problems now and look forward to your help.
For baselines Drnet, Tarnet and VCnet, there is a big difference between the result in Table 5 and my reproduced one:
This attached figure is the results on test dataset and validation dataset under Drnet:
As for as I know, the last indicator represents AMSE and in different simulation experiments, the results are similar.
So, I wonder are there some hyper-parameters should be tuned for these three baselines?
For your proposed TransTEE, I can obtain a similar result as you provided in Table 5 by running the command:
python test_TransTEE.py --num_treatments=3 --treatment_selection_bias=2.0 --dosage_selection_bias=2.0 --batch_size=128 --h_dim=64 --h_inv_eqv_dim=64 --num_dosage_samples=5 --alpha=1.0 --model_name="TransTEE"
But if I tune the hyper-parameters based on Table 8, the results are much worse.
Are the parameter values in Table 8 not suitable for TCGA dataset in this scenario?
Hi, your work is so great and thanks for your sharing!
I have a question about the Dosage/datas_simulation/get_split_indices fucntion (Line 131~138). The rest_indices should be split into train_indices and val_indices, but the values in train_indices and val_indices are limited to the range of [0, len(rest_indices)], other than the values in rest_indices. In that case, the train_indices and val_indices don't overlap, but they are overlapped with test_indices.
If there is anything wrong with my understanding, please point out. Thanks!
In TransTEE/Continuous/models there is no file DisCri, so the main_batch.py cannot be run. Could this file please be added?
Hi,
Based on your detailed intructions, I try to reproduce the results on IHDP datasets and have some problems.
Under the setting Extrapolation (h=2), the results of TransTEE in Table 2 is 0.2745+/-0.14976, while I get 0.4783+/-0.3629. My results are aslo computed with 100 datasets under default parameters.
Then I adjust the epochs from the default 1500 to 2000 and 2500, the results are 0.3418+/-0.3019 and 0.2551+/-0.2370 separately. Although 0.2551 is close to your result, the standard deviation is much larger than yours.
In that case, could you give me some guidance about the result reproduction?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.