Adversarial Precision Sensing

This is the code repository for Adversarial Precision Sensing with Healthcare Applications (ICDM'20) by Fengyi Tang, Lifan Zeng, Fei Wang and Jiayu Zhou. Here, you will find instructions for:

• preparing MIMIC-III code to test out experiments
• preparing synthetic data from the paper
• models.py file containing the FS and PL models
• main.py training and evaluation loops

Usages

Requirements

• Python 3.4+
• Pytorch (https://pytorch.org/)
• Scikit-Learn
• Gensim
• NumPy
• Matplotlib
• Seaborn
• Pandas
• Tensorflow 1.11+
• Progressbar2
• Postgres (or equivalent for building local MIMIC-III)

MIMIC-III

Please apply for access to the publicly available MIMIC-III DataBase via https://www.physionet.org/.

Files

• utils.py: utilities file containing visualization methods, variable conversion to GPU compatibility form, subsampling, ... etc.
• models.py: contains all the model classes for inference models as well as the FS and PL.
• preprocessing.py: preprocesses local MIMIC-III data into X and y.
• main.py: runs the main training loop, testing, and trade-off studies.

Instructions for Use

Workflow: MIMIC-III Access -> Obtain Views and Tables -> Preprocessing -> Pipeline

1. Obtain access to MIMIC-III and clone this repo to local folder. Create a local MIMIC-III folder to store a few files:

• .../local_mimic
• .../local_mimic/views
• .../local_mimic/tables
• .../local_mimic/save

These paths will be important for storing views and pivot tables, which will be used for preprocessing.

2. Build MIMIC-III database using postgres, follow the instructions outlined in the MIMIC-III repository: https://github.com/MIT-LCP/mimic-code/tree/master/buildmimic/postgres.

3. Go to the pivot folder in the MIMIC-III repository: https://github.com/MIT-LCP/mimic-code/tree/master/concepts/pivot. Run use the .sql scripts to build a local set of .csv files of the pivot tables:

• pivoted-bg.sql
• pivoted_vital.sql
• pivoted_lab.sql
• pivoted_gcs.sql (optional)
• pivoted_uo.sql (optional)

When running the .sql script, change the delimiter of the materialized views to ',' when saving as .csv file.

For example,
mimic=> \copy (select * FROM mimiciii.icustay_detail) to 'icustay_detail.csv' delimiter ',' csv header;

After running these scripts, you should have obtained local .csv files of the pivot tables. Create a local folder to place them in, i.e. .../local_mimic/views/pivoted-bg.csv. Remember this .../local_mimic/views folder, as it will be the path_views input for preprocessing purposes.

4. Go to the demographics folder in the MIMIC-III repository: https://github.com/MIT-LCP/mimic-code/tree/master/concepts/demographics.

Run icustay-detail.sql and obtain a local .csv file of icustays-detail view. Create a local folder to place the .csv file in, i.e..../local_mimic/views/icustay_details.csv. Again, have this .csv file inside the local views folder.

A minor change needs to be made in icustay_details.csv:
change 'admission_age' -> 'age' for the column header in the .csv file manually.

5. Obtain a local copy of the following tables from MIMIC-III:

• admissions.csv
• diagnoses_icd.csv
• d_icd_diagnoses.csv

These can be directly obtained from Physionet as compressed files. While tables such as chartevents are large, the above tables are quite small and easy to query directly if a local copy is available.

Save these tables under .../local_mimic/tables folder. Make the following changes:

• In ~/local_mimic/tables/diagnoses_icd.csv, change the column titles "ROW_ID","SUBJECT_ID","HADM_ID","SEQ_NUM","ICD9_CODE" to "row_id","subject_id","hadm_id","seq_num","icd9_code" (i.e., make lower case).
• In ~local_mimic/tables/d_icd_diagnoses.csv change the column titles "ROW_ID","ICD9_CODE","SHORT_TITLE","LONG_TITLE" to "row_id","icd9_code","short_title","long_title" (i.e., again, make lower case).

6. Run preprocessing.py with inputs:

• --path_tables <path_tables>
• --path_views <path_views>
• --path_save <path_save>.

<path_tables> and <path_views> should correspond to the folders under which the local tables and views (pivots and icustays-details) are saved. <path_save> corresponds to the desired folder to save your variables for training and beyond.

preprocessing.py will generate the following files:

• X.npy: main feature tensor, consisting of time-series data generated from a combination of 19 lab values and vital signs over 48 hour period from start of admissions.
• labels: raw file containing hadm_id (hospital admissions identifier) and labels of interest.
• y: main label matrix with mortality label for each patient.

7. Run main.py with selection of task, training, and testing conditions:

• --features_dir: path to saved the feature file to use as X. Selections include X19, X48, sentences, or onehot.
• --y_dir: path to y.
• --task: specifies the learning task. User can choose between ['mort', 'sep'] (default = mort).
• --budget: specifies budget to be used for main training and testing loops (default=1e-5).
• --checkpoint_dir: specifies the path to save best models and testing results.
• --hidden_size: specifies number of hidden units for deep models (default=128).
• --l1_reg: regularization for inference models (default=5e-6)
• --learning_rate: specifies the initial learning rate (default=0.005).
• --training_epochs: number of training epochs for inference models (default=45).
• --num_epochs: max number of training epochs for FS and PL (default=30). *--batch_size: batch size during training (default = 32).

The main program runs the following items:

• 5 evaluation runs of the inference models on the sensed vs. original data.
• Sparsity trade-off experiments. The default beta is set to 5e-6, which yields the best results but can be unstable during training. The main function returns stats and beta_data for each of the experiments and saves stats (main experiments). Models are saved under checkpoint_dir provided by the user.

8. Instructions for running Synthetic Experiments.

• First, download the dataset synthetic.data.gz from https://archive.ics.uci.edu/ml/machine-learning-databases/synthetic-mld/.
• Unzip the synthetic.data.gz to obtain synthetic.data file. Save this file under some desired path, i.e, filename.
• Run syn_exp(filename), with filename being the path to the saved .data file.
• syn_exp returns:
  • data: generated synthetic (X,y) pairs used for the experiments.
  • stats: dictionary of the form {'data': performance, 'budget': % sparsity in C, 'G': parameters of FS, 'D': parameters of PL}
  • mask: BCPS sensed features.
  • ground_truth: true C.

In this version, one has to manually change the sparsity patterns in the ground truth C. Go to X[:, xx:yy, zz:aa] and ground_truth[:, xx:yy, zz:aa] and change the xx:yy to the desired time-steps and zz:aa to be the desired features that contribute to the underlying y label. Similarly, in the y label loop, change if (np.sum(X[i][xx:yy, zz:aa]) + np.sum(...) ... > 0) to the same set of xx:yy and zz:aa for ground_truth.

To visualize the sensed features, use visualize_mask(mask[i]) for the i-th test sample.

References

If you find this repository helpful for your work, please consider citing us.