unpack zip file and delete zip file

demo_code/compute_snoozed_auc.py

@@ -0,0 +1,142 @@
+# Copyright 2021
+# Supreeth P. Shashikumar <spshashikumar@health.ucsd.edu>
+# Shamim Nemati <snemati@health.ucsd.edu>
+# Revision: 1.1 Date: 08/03/2021
+
+import numpy as np
+
+def compute_confusion_matrix_per_patient(labels, predictions, threshold, seqLength, includeMaskPerPatient, false_positive_prediction_horizon = 44, snooze_length = 6):   
+    # Fill-in confusion matrix.
+    tp = 0
+    fp = 0
+    fn = 0
+    tn = 0
+
+    tp_modified = 0
+    fp_modified = 0
+
+    snooze = 0
+
+    for t in range(seqLength):       
+        # Only check labels and predictions when not snoozed.
+        if snooze == 0:           
+            if includeMaskPerPatient[t]==1:
+                label = labels[t] == 1
+                prediction = predictions[t] >= threshold 
+
+                # Is there a positive label within the horizon?
+                label_horizon = np.any(labels[t:min(seqLength, t+false_positive_prediction_horizon+1)])
+
+                if label_horizon and prediction:
+                    tp += 1
+                elif not label_horizon and prediction:
+                    fp += 1
+                elif label and not prediction:
+                    fn += 1
+                elif not label and not prediction:
+                    tn += 1
+
+                if prediction:
+                    snooze = snooze_length
+
+        else:
+            snooze -= 1
+
+    return tp, fp, fn, tn
+
+
+def compute_snoozed_auc(labels, predictions,  seqLength, includeMaskPerPatient, false_positive_prediction_horizon=44, snooze_length = 6, num_thresholds = 500):
+    # Check input for errors.
+    if len(labels) != len(predictions):
+        raise Exception('Numbers of labels and predictions must be equal.')
+    if any(len(X) != len(Y) for X, Y in zip(labels, predictions)):
+        raise Exception('Numbers of labels and predictions must be equal for all samples.')
+    if any(not (x == 0 or x == 1) for X in labels for x in X):
+        raise Exception('Labels must be binary.')
+    if any(not (0 <= y <= 1) for Y in predictions for y in Y):
+        raise Exception('Predictions must be between 0 and 1, inclusive.')
+
+    # Find prediction thresholds.
+    #num_cohorts = len(labels)
+    sorted_predictions = [np.unique(Y) for Y in predictions]
+
+    if not num_thresholds:
+        thresholds = np.unique(np.concatenate(sorted_predictions))[::-1]
+    else:
+        percentiles = np.linspace(0, 100, num_thresholds)
+        thresholds = np.unique(np.percentile(np.concatenate(sorted_predictions), percentiles, interpolation='nearest'))[::-1]
+
+    if thresholds[0] != 1:
+        thresholds = np.insert(thresholds, 0, 1)
+    if thresholds[-1] == 0:
+        thresholds = thresholds[:-1]
+    num_thresholds = len(thresholds)
+
+    num_patients = labels.shape[0]
+
+    # Populate confusion matrix for each prediction threshold.
+    sample_tp = np.zeros(num_patients)
+    sample_fp = np.zeros(num_patients)
+    sample_fn = np.zeros(num_patients)
+    sample_tn = np.zeros(num_patients)
+    sample_indices = -np.ones(num_patients, dtype=np.int)
+
+    tp = np.zeros(num_thresholds)
+    fp = np.zeros(num_thresholds)
+    fn = np.zeros(num_thresholds)
+    tn = np.zeros(num_thresholds)
+
+    for i in range(num_thresholds):
+        for j in range(num_patients):
+            k = np.searchsorted(sorted_predictions[j], thresholds[i])
+            if k != sample_indices[j]:
+                a, b, c, d = compute_confusion_matrix_per_patient(labels[j], predictions[j], thresholds[i], seqLength[j], includeMaskPerPatient[j], false_positive_prediction_horizon, snooze_length)
+                sample_tp[j] = a
+                sample_fp[j] = b
+                sample_fn[j] = c
+                sample_tn[j] = d
+                sample_indices[j] = k
+
+            tp[i] += sample_tp[j]
+            fp[i] += sample_fp[j]
+            fn[i] += sample_fn[j]
+            tn[i] += sample_tn[j]
+
+    tpr = np.zeros(num_thresholds)
+    tnr = np.zeros(num_thresholds)
+    ppv = np.zeros(num_thresholds)
+    npv = np.zeros(num_thresholds)
+
+    for i in range(num_thresholds):
+        if tp[i] + fn[i]:
+            tpr[i] = tp[i] / (tp[i] + fn[i])
+        else:
+            tpr[i] = 1
+        if fp[i] + tn[i]:
+            tnr[i] = tn[i] / (fp[i] + tn[i])
+        else:
+            tpr[i] = 0
+        if tp[i] + fp[i]:
+            ppv[i] = tp[i] / (tp[i] + fp[i])
+        else:
+            ppv[i] = 1
+
+        if fn[i] + tn[i]:
+            npv[i] = tn[i] / (fn[i] + tn[i])
+        else:
+            npv[i] = 0
+
+    # Compute AUROC as the area under a piecewise linear function of TPR
+    # (x-axis) and FPR (y-axis) and AUPRC as the area under a piecewise linear
+    # function of recall (x-axis) and precision (y-axis).
+    auroc = 0
+    auprc = 0
+    auprc_modified = 0
+    for i in range(num_thresholds - 1):
+        auroc += 0.5 * (tpr[i + 1] - tpr[i]) * (tnr[i + 1] + tnr[i])
+        auprc += (tpr[i + 1] - tpr[i]) * (ppv[i + 1])
+
+#     for i in range(num_thresholds):
+#         print(i, thresholds[i], tp[i], fp[i], fn[i], tn[i])
+
+    return auroc, auprc, tpr, tnr, fp, ppv, thresholds, npv

demo_code/demo_data/p00018.csv

@@ -0,0 +1,20 @@
+3.524158596992492676e-01,0,0
+2.373938113451004028e-01,0,0
+1.946674138307571411e-01,1,0
+3.095490932464599609e-01,0,0
+2.915228009223937988e-01,0,0
+2.742238640785217285e-01,0,0
+1.463383585214614868e-01,1,0
+1.215671002864837646e-01,1,0
+1.064075678586959839e-01,1,0
+9.595078974962234497e-02,1,0
+8.819054067134857178e-02,1,0
+6.943967193365097046e-02,1,0
+7.014473527669906616e-02,1,0
+7.069533318281173706e-02,1,0
+7.114680856466293335e-02,1,0
+1.151653826236724854e-01,1,0
+1.148346960544586182e-01,1,0
+1.156225427985191345e-01,1,0
+1.163050234317779541e-01,1,0
+1.481731981039047241e-01,1,0

demo_code/demo_data/p00056.csv

@@ -0,0 +1,21 @@
+8.673850446939468384e-02,1,0
+7.706475257873535156e-02,1,0
+4.691105708479881287e-02,1,0
+1.722965985536575317e-01,0,0
+1.190305575728416443e-01,1,0
+6.667682528495788574e-02,1,0
+4.770652949810028076e-02,1,0
+7.717213034629821777e-02,1,0
+7.408756017684936523e-02,1,0
+1.183828860521316528e-01,1,0
+5.475239828228950500e-02,1,0
+2.908693067729473114e-02,1,0
+2.977270446717739105e-02,1,0
+3.379622846841812134e-02,1,0
+2.765891142189502716e-02,1,0
+3.352526575326919556e-02,1,0
+2.683576196432113647e-02,1,0
+2.441511675715446472e-02,1,0
+2.447645366191864014e-02,1,0
+3.150004893541336060e-02,1,0
+3.206261992454528809e-02,1,0

demo_code/demo_data/p00062.csv

@@ -0,0 +1,24 @@
+2.327396124601364136e-01,0,0
+1.799809336662292480e-01,0,0
+1.609969884157180786e-01,0,0
+1.109594926238059998e-01,0,0
+9.771641343832015991e-02,0,0
+1.133971586823463440e-01,0,0
+1.335569769144058228e-01,0,0
+1.050397530198097229e-01,1,0
+1.781912744045257568e-01,0,0
+3.535856008529663086e-01,0,0
+1.905256658792495728e-01,1,0
+1.901751309633255005e-01,1,0
+2.039649486541748047e-01,1,0
+3.655618429183959961e-01,0,0
+1.842548698186874390e-01,1,0
+3.691251873970031738e-01,0,0
+2.257111072540283203e-01,1,0
+3.855168223381042480e-01,0,0
+7.639527916908264160e-01,1,0
+3.671570420265197754e-01,0,0
+3.083380460739135742e-01,0,0
+3.002890348434448242e-01,0,0
+4.024444222450256348e-01,0,0
+3.837247490882873535e-01,0,0

demo_code/demo_data/p00090.csv

@@ -0,0 +1,26 @@
+2.488320022821426392e-01,0,0
+1.979351341724395752e-01,1,0
+1.607611626386642456e-01,1,0
+1.380303651094436646e-01,1,0
+1.223617047071456909e-01,1,0
+1.107598841190338135e-01,1,0
+7.814987748861312866e-02,1,0
+7.159715890884399414e-02,1,0
+5.468975380063056946e-02,1,0
+5.159991607069969177e-02,1,0
+4.954269900918006897e-02,1,0
+4.854648187756538391e-02,1,0
+4.865250363945960999e-02,1,0
+4.899834841489791870e-02,1,0
+5.747842416167259216e-02,1,0
+5.782412737607955933e-02,1,0
+6.699778139591217041e-02,1,0
+6.522141396999359131e-02,1,0
+6.467941403388977051e-02,1,0
+6.444168835878372192e-02,1,0
+6.435013562440872192e-02,1,0
+6.805702298879623413e-02,1,0
+6.775090098381042480e-02,1,0
+6.776337325572967529e-02,1,0
+6.632813811302185059e-02,1,0
+5.466015264391899109e-02,1,0

demo_code/demo_data/p00101.csv

@@ -0,0 +1,21 @@
+6.220677495002746582e-01,1,0
+5.073778629302978516e-01,1,0
+5.705167651176452637e-01,1,0
+4.508942365646362305e-01,1,0
+3.859788179397583008e-01,1,0
+3.305697739124298096e-01,0,0
+2.618824243545532227e-01,0,0
+6.680250763893127441e-01,1,0
+5.181686282157897949e-01,1,0
+3.212387263774871826e-01,0,0
+3.445956110954284668e-01,0,0
+4.637837111949920654e-01,1,0
+1.282385736703872681e-01,1,0
+3.278776705265045166e-01,0,0
+4.151531159877777100e-01,1,0
+3.990756273269653320e-01,1,0
+4.129233062267303467e-01,1,0
+4.006917178630828857e-01,1,0
+1.480282992124557495e-01,1,0
+1.325468420982360840e-01,1,0
+2.799977362155914307e-01,0,0

demo_code/demo_data/p00104.csv

@@ -0,0 +1,38 @@
+1.961058080196380615e-01,0,0
+1.987523287534713745e-01,0,0
+2.982433438301086426e-01,0,0
+2.951050996780395508e-01,0,0
+2.228471338748931885e-01,0,0
+4.143873453140258789e-01,0,0
+2.285730093717575073e-01,0,0
+3.676817417144775391e-01,0,0
+2.382458299398422241e-01,0,0
+2.074865549802780151e-01,0,0
+2.154600173234939575e-01,0,0
+2.412652522325515747e-01,0,0
+2.622527778148651123e-01,0,0
+2.347321957349777222e-01,1,0
+5.215241909027099609e-01,1,0
+3.982219398021697998e-01,0,0
+1.449419856071472168e-01,1,0
+3.473561704158782959e-01,0,0
+1.706304848194122314e-01,1,0
+3.519901037216186523e-01,0,0
+2.256698608398437500e-01,0,0
+1.722945570945739746e-01,1,0
+2.874557077884674072e-01,0,0
+4.335812330245971680e-01,1,0
+3.675619959831237793e-01,0,0
+3.314910829067230225e-01,0,0
+4.179086685180664062e-01,1,0
+6.239072680473327637e-01,1,0
+6.107838749885559082e-01,1,0
+3.462873697280883789e-01,0,0
+3.909673392772674561e-01,0,0
+2.948751449584960938e-01,0,0
+6.219827532768249512e-01,1,0
+4.987363815307617188e-01,1,0
+6.992628574371337891e-01,1,0
+6.030540466308593750e-01,1,0
+6.890112161636352539e-01,1,0
+4.800139367580413818e-01,1,0

demo_code/demo_data/p00111.csv

@@ -0,0 +1,87 @@
+8.455271273851394653e-02,0,0
+4.950173199176788330e-01,1,0
+4.740326702594757080e-01,0,0
+6.111966371536254883e-01,1,0
+4.562445580959320068e-01,0,0
+2.932156920433044434e-01,0,0
+2.848603129386901855e-01,0,0
+1.630455851554870605e-01,0,0
+2.662973403930664062e-01,0,0
+3.305304348468780518e-01,0,0
+3.301716446876525879e-01,0,0
+2.828376293182373047e-01,0,0
+2.870158851146697998e-01,0,0
+2.684437632560729980e-01,0,0
+2.781126797199249268e-01,0,0
+3.532216846942901611e-01,0,0
+3.163609504699707031e-01,0,0
+3.460981547832489014e-01,0,0
+2.954213917255401611e-01,0,0
+5.443381071090698242e-01,1,0
+5.360348224639892578e-01,1,0
+4.693817496299743652e-01,1,0
+6.658125519752502441e-01,1,0
+6.294396519660949707e-01,1,0
+7.849550843238830566e-01,1,0
+8.144682645797729492e-01,1,0
+7.429504394531250000e-01,1,0
+6.779129505157470703e-01,1,0
+5.961812734603881836e-01,1,0
+7.158362865447998047e-01,1,0
+6.835216283798217773e-01,1,0
+6.447662115097045898e-01,1,0
+6.715291142463684082e-01,1,0
+4.188896715641021729e-01,0,0
+4.550487995147705078e-01,1,0
+5.948814153671264648e-01,1,0
+6.547959446907043457e-01,1,0
+5.404837727546691895e-01,1,0
+6.065972447395324707e-01,1,0
+7.805557250976562500e-01,1,0
+8.571919202804565430e-01,1,0
+6.618452668190002441e-01,1,0
+7.575771212577819824e-01,1,0
+6.720529794692993164e-01,1,0
+5.791990756988525391e-01,1,0
+5.727471113204956055e-01,1,0
+7.733771204948425293e-01,1,0
+6.616895794868469238e-01,1,0
+6.000316739082336426e-01,1,0
+6.993266940116882324e-01,1,0
+6.265998482704162598e-01,1,0
+7.655912637710571289e-01,1,0
+6.127461791038513184e-01,1,0
+5.661484599113464355e-01,1,0
+6.401023864746093750e-01,1,0
+5.669670701026916504e-01,1,0
+6.258417367935180664e-01,1,0
+6.392814517021179199e-01,1,0
+6.246948242187500000e-01,1,0
+6.919012665748596191e-01,1,0
+5.874356031417846680e-01,1,0
+6.200698614120483398e-01,1,0
+5.449547171592712402e-01,1,0
+7.387002110481262207e-01,1,0
+7.545434236526489258e-01,1,0
+7.274111509323120117e-01,1,0
+6.821124553680419922e-01,1,0
+6.572970151901245117e-01,1,0
+6.780682206153869629e-01,1,0
+7.428504228591918945e-01,1,0
+7.313720583915710449e-01,1,0
+7.665268778800964355e-01,1,0
+7.065571546554565430e-01,1,0
+6.811770200729370117e-01,1,0
+7.090263962745666504e-01,1,0
+7.107076048851013184e-01,1,0
+6.129837632179260254e-01,1,0
+7.160031795501708984e-01,1,0
+5.195250511169433594e-01,1,0
+4.911837577819824219e-01,1,0
+6.014863848686218262e-01,1,0
+5.876170992851257324e-01,1,0
+6.298850178718566895e-01,1,0
+7.663446664810180664e-01,1,0
+7.325693964958190918e-01,1,0
+7.227794528007507324e-01,1,0
+6.508201956748962402e-01,1,0