--- a
+++ b/preprocessing/4-combine.py
@@ -0,0 +1,97 @@
+import json
+import sparse
+import pandas as pd
+import numpy as np
+import scipy.sparse
+import joblib
+
+def load_IDs(fname):
+    IDs = pd.read_csv(fname, header=0, names=['ID'])
+    IDs.index.name = 'i'
+    IDs = IDs.reset_index()
+    return IDs
+
+def _get_feature_set(df, X_ALL, IDs_ALL):
+    IDs = df.set_index('ID')[[]]
+    idx = IDs.join(IDs_ALL.set_index('ID')).astype(float)
+    X = [X_ALL[int(i),:] if not np.isnan(i) else sparse.zeros(X_ALL.shape[1]) for i in idx.values]
+    return sparse.stack(X)
+
+def get_features(df, feature_sets):
+    features = []
+    feature_names = []
+    if 'demog' in feature_sets:
+        X_d = df.set_index('hosp_id')[['ID']].join(df_demog).reset_index(drop=True).set_index('ID').loc[df['ID']]
+        X_d = sparse.as_coo(X_d.values)
+        features.append(X_d)
+        feature_names.append(names_demog)
+        print('demog - Done')
+    if 'vitals' in feature_sets:
+        X_v = _get_feature_set(df, X_vitals, IDs_vitals)
+        features.append(X_v)
+        feature_names.append(names_vitals)
+        print('vitals - Done')
+    if 'meds' in feature_sets:
+        X_m = _get_feature_set(df, X_meds, IDs_meds)
+        features.append(X_m)
+        feature_names.append(names_meds)
+        print('meds - Done')
+    if 'labs' in feature_sets:
+        X_l = _get_feature_set(df, X_labs, IDs_labs)
+        features.append(X_l)
+        feature_names.append(names_labs)
+        print('labs - Done')
+    if 'flow' in feature_sets:
+        print('flow', end='')
+        X_f = _get_feature_set(df, X_flow, IDs_flow)
+        features.append(X_f)
+        feature_names.append(names_flow)
+        print(' - Done')
+    X = sparse.concatenate(features, axis=1).tocsr()
+    feature_names = sum(feature_names, [])
+    return X, np.array(feature_names)
+
+
+if __name__ == '__main__':
+
+    df_demog = pd.read_csv('sample_output/out_demog/static-features.csv').set_index('hosp_id')
+    names_demog = list(df_demog.columns)
+    print('demog - Loaded')
+
+    X_vitals = sparse.load_npz('sample_output/out_vitals/X_all.npz')
+    IDs_vitals = load_IDs('sample_output/out_vitals/X_all.IDs.csv')
+    names_vitals = json.load(open('metadata/vitals/X_all.feature_names.json', 'r'))
+    print('vitals - Loaded')
+
+    X_meds = sparse.load_npz('sample_output/out_meds/X_all.npz')
+    IDs_meds = load_IDs('sample_output/out_meds/X_all.IDs.csv')
+    names_meds = json.load(open('metadata/meds/X_all.feature_names.json', 'r'))
+    print('meds - Loaded')
+
+    X_labs = sparse.load_npz('sample_output/out_labs/X_all.npz')
+    IDs_labs = load_IDs('sample_output/out_labs/X_all.IDs.csv')
+    names_labs = json.load(open('metadata/labs/X_all.feature_names.json', 'r'))
+    print('labs - Loaded')
+
+    X_flow = sparse.load_npz('sample_output/out_flow/X_all.npz')
+    IDs_flow = load_IDs('sample_output/out_flow/X_all.IDs.csv')
+    names_flow = json.load(open('metadata/flow/X_all.feature_names.json', 'r'))
+    print('flow - Loaded')
+
+    df_cohort = pd.read_csv('sample_input/windows_map.csv')
+    X, names = get_features(df_cohort, ['demog', 'vitals', 'meds', 'labs', 'flow'])
+    df_features = pd.DataFrame(X.todense(), columns=names, index=df_cohort['ID'])
+    pd.Series(names).rename('feature_name').to_csv('./sample_output/feature_names.csv', index=False)
+
+    ## Full feature matrix
+    joblib.dump(df_features, 'sample_output/full.joblib')
+
+    ## Baseline features
+    baseline_cols = pd.read_csv('metadata/Baseline_Feature_Names.txt', sep='\t', header=None)[0].values
+    df_baseline = df_features[baseline_cols]
+    df_baseline.to_csv('sample_output/baseline.csv')
+
+    ## M-CURES (lite)
+    mcures_cols = pd.read_csv('metadata/MCURES_Feature_Names.txt', sep='\t', header=None)[0].values
+    df_mcures = df_features[mcures_cols]
+    df_mcures.to_csv('sample_output/mcures.csv')