Skip to content

Test 09 - Features With Missing Values Analysis

Purpose

Run feature analysis that specifically inspects missingness patterns for a selected set of important features. Computes statistics and plots for features where missing values are informative or prevalent.

Required Inputs

From configs/env.sh and the test invocation environment:

  • WORK_DIR: Working folder where outputs will be written
  • FIRST_WORK_DIR: Reference run (contains model and rep)
  • MODEL_PATH: Path to the model to be used for generating features
  • CONFIG_DIR: Directory holding feat_resolution.tsv and similar config files
  • CMP_FEATURE_RES: Comma-separated list of feature:resolution tokens used to define the top features and per-feature resolution

How to Run

From your TestKit folder, execute:

1
./run.specific.sh 9

Or include as part of the full suite:

1
./run.sh

Outputs are placed under ${WORK_DIR}/outputs/features_stats.with_missings.tsv and ${WORK_DIR}/outputs/graphs_with_missings.

What This Test Does

  • Loads the repository with med.PidRepository() and initializes it with ${WORK_DIR}/rep/test.repository.
  • Loads the model from ${FIRST_WORK_DIR}/model/model.medmdl, fits it to the repository, and applies feature generators to the samples (${WORK_DIR}/Samples/3.test_cohort.samples), producing a feature DataFrame.
  • Selects top_features from CMP_FEATURE_RES and applies per-feature resolution adjustments.
  • For each feature:
    • Computes overall, case, and control means, stds and missingness percentages (missing is expected to be coded as -65336).
    • Writes results to ${WORK_DIR}/outputs/features_stats.with_missings.tsv.
    • Generates plots (markers+lines) saved under ${WORK_DIR}/outputs/graphs_with_missings (plotly HTML files).
    • Computes KLD between case and control distributions and prints results.

Output Location

  • Stats TSV: ${WORK_DIR}/outputs/features_stats.with_missings.tsv
  • Graphs: ${WORK_DIR}/outputs/graphs_with_missings (one HTML per feature)

How to Interpret Results

  • Use the TSV to quickly see which features have high missingness and how that differs between cases and controls.
  • Open HTML graphs to inspect distribution patterns and whether missingness behaves differently across outcome groups.
  • Elevated KLD suggests the feature's distribution differs notably between cases and controls (including missingness patterns).

Common failure modes and suggestions

  • Failure to load repository or model:
    • Ensure ${WORK_DIR}/rep/test.repository and ${FIRST_WORK_DIR}/model/model.medmdl exist.
  • Feature name mismatches or multiple substring matches:
    • The script matches features by substring and will fail if it finds zero or multiple matches. Confirm naming conventions.
  • Missing med Python package or model runtime dependencies:
    • Ensure the runtime Python environment has the med package and project helpers installed.

Notes and Implementation Details

  • Feature resolution is applied by rounding values to the specified resolution in CMP_FEATURE_RES.

Test Results Review

Primary files to inspect after running this test:

  • ${WORK_DIR}/outputs/features_stats.with_missings.tsv
  • ${WORK_DIR}/outputs/graphs_with_missings/*.html