# Genentech – 404 Challenge — Data

来源：https://www.kaggle.com/competitions/genentech-404-challenge/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

## Data Access
All participants are required to sign and adhere to the Data Usage Agreement for the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset on the LONI website [https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI](https://ida.loni.usc.edu/collaboration/access/appApply.jsp?project=ADNI). Participants must request access to the database, be aware that approval may take between 3-5 business days. Participants who already have approval to access ADNI through the LONI website, do not need to apply again. PARTICIPANTS SHOULD NOT DOWNLOAD THE COMPETITION DATA UNTIL THE ADNI DATA ACCESS REQUEST HAS BEEN APPROVED BY THE LONI WEBSITE ADMINISTRATORS. 

In the Proposed Analysis section, Participants are encouraged to use the following statement:
> I am requesting access to the ADNI database in order to participate in Genentech's 404 Challenge on missing data imputation. The goal of the 404 Challenge to develop flexible algorithms to deal with different patterns of missing data in tabular clinical datasets that include demographics, cognitive scores, and MR-derived volumetric measurements.

## Files
The training set (dev_set.csv) is a complete dataset with no missing values. Three copies of dev_set.csv are created, dev_1.csv, dev_2.csv, dev_3.csv each with different simulated missingness mechanisms. 

The task is to predict the missing values in the test_A.csv and test_B.csv. The solution file should be formatted similarly to the sample_submission.csv. The Id column should include a string with RID_hash , VISCODE ,  col_name ,   test_set separated by underscores, for example `7cdf3031d8e8b148faf18742c72e6f21ffd63e9a344f729b9f33c7bd95e69355_90_CDRSB_test_B`. The Predicted column should include the predicted value.

## Files

*   **dev_set.csv** - complete training set
*   **dev_1.csv** - training set with missing values, missingness mechanism #1
*   **dev_2.csv** - training set with missing values, missingness mechanism #2
*   **dev_3.csv** - training set with missing values, missingness mechanism #3

*   **test_A.csv** - test set with missing values, missingness mechanism A
*   **test_B.csv** - test set with missing values, missingness mechanism B

*   **sample_submission.csv** - a sample submission file in the correct format, predicting all missing values as the column means

*Hint*: Participants will not be given information about missingness mechanisms A, B but are encouraged to develop algorithms that can characterize the missingness pattern and adapt the imputation approach accordingly.

## Leaderboards
Only two submissions to the leaderboard are allowed per day. The public leaderboard contains 20% of all test_A samples and 20% of all test_B samples. The private leaderboard contains the remaining 80% of each dataset. Only the private leaderboard will be used to determine the challenge winner.

## Columns
Each row in the csv corresponds to a single subject visit. The data is longitudinal and all datasets (training set, public leaderboard set, and private leaderboard set) were split by subject. Ordinal or categorical variables will be treated as continuous during evaluation. Performance will be measured using range normalized mean absolute error (MAE).

Variables that are always observed:
*   `RID_hash` - unique subject ID
*   `VISCODE ` - visit code, referring to the number of months

Variables that may have missingness:
*   `AGE ` - age at that particular visit
*   `PTGENDER_num ` - sex {0: 'Male', 1: 'Female'}
*   `PTEDUCAT ` - education level
*   `DX_num ` - diagnosis {0: 'Cognitively Normal', 1: 'Mild Cognitive Impairment', 2: 'Dementia'}
*   `APOE4 ` - number of APOE e4 alleles
*   `CDRSB ` - cognitive score, Clinical Dementia Rating Sum of Boxes 
*   `MMSE ` - cognitive score, Mini‐Mental State Examination
*   `ADAS13 ` - cognitive score, Alzheimer's Disease Assessment Scale-Cognitive Subscale
*   `Ventricles ` - ventricle volume 
*   `Hippocampus ` - hippocampus volume 
*   `WholeBrain ` - whole brain volume
*   `Entorhinal ` - entorhinal volume
*   `Fusiform ` - fusiform volume
*   `MidTemp ` - midtemp volume