

In [2]:

    
%matplotlib inline
%load_ext autoreload
%autoreload 2

from __future__ import division
from __future__ import print_function
from __future__ import absolute_import
from __future__ import unicode_literals  

import math
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import sklearn.linear_model as linear_model

import scipy
import sklearn

sns.set(color_codes=True)

import tensorflow as tf
from tensorflow.contrib.learn.python.learn.datasets import base

from influence.binaryLogisticRegressionWithLBFGS import BinaryLogisticRegressionWithLBFGS
from influence.smooth_hinge import SmoothHinge
import influence.dataset as dataset
from influence.dataset import DataSet

np.random.seed(42)









    



/juicier/scr100/scr/pangwei/influence_release
The autoreload extension is already loaded. To reload it, use:
  %reload_ext autoreload






    



Using TensorFlow backend.



In [3]:

    
def examine_vec(x, verbose=False):
    assert len(feature_names) == len(x)
    print('Age: %s' % x[age_var_indices])
    if verbose:
        for feature_name, val in zip(feature_names, x):
            print('%32s: %.6f' % (feature_name, val))
    
def examine_train_point(idx, verbose=False):
    print('Label: %s' % Y_train[idx])
    examine_vec(modified_X_train[idx, :], verbose)
    
def examine_test_point(idx, verbose=False):
    print('Label: %s' % Y_test[idx])
    examine_vec(X_test[idx, :], verbose)

Read and process dataset



In [5]:

    
df = pd.read_csv('diabetic_data.csv')
# Use this if you are not running this in CodaLab
# df = pd.read_csv('../data/diabetic_data.csv')



In [7]:

    
df









    Out[7]:






  
    
      
      encounter_id
      patient_nbr
      race
      gender
      age
      weight
      admission_type_id
      discharge_disposition_id
      admission_source_id
      time_in_hospital
      ...
      citoglipton
      insulin
      glyburide-metformin
      glipizide-metformin
      glimepiride-pioglitazone
      metformin-rosiglitazone
      metformin-pioglitazone
      change
      diabetesMed
      readmitted
    
  
  
    
      0
      2278392
      8222157
      Caucasian
      Female
      [0-10)
      ?
      6
      25
      1
      1
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      NO
    
    
      1
      149190
      55629189
      Caucasian
      Female
      [10-20)
      ?
      1
      1
      7
      3
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      2
      64410
      86047875
      AfricanAmerican
      Female
      [20-30)
      ?
      1
      1
      7
      2
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      3
      500364
      82442376
      Caucasian
      Male
      [30-40)
      ?
      1
      1
      7
      2
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      4
      16680
      42519267
      Caucasian
      Male
      [40-50)
      ?
      1
      1
      7
      1
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      5
      35754
      82637451
      Caucasian
      Male
      [50-60)
      ?
      2
      1
      2
      3
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      6
      55842
      84259809
      Caucasian
      Male
      [60-70)
      ?
      3
      1
      2
      4
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      7
      63768
      114882984
      Caucasian
      Male
      [70-80)
      ?
      1
      1
      7
      5
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      8
      12522
      48330783
      Caucasian
      Female
      [80-90)
      ?
      2
      1
      4
      13
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      9
      15738
      63555939
      Caucasian
      Female
      [90-100)
      ?
      3
      3
      4
      12
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      10
      28236
      89869032
      AfricanAmerican
      Female
      [40-50)
      ?
      1
      1
      7
      9
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      11
      36900
      77391171
      AfricanAmerican
      Male
      [60-70)
      ?
      2
      1
      4
      7
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      <30
    
    
      12
      40926
      85504905
      Caucasian
      Female
      [40-50)
      ?
      1
      3
      7
      7
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      <30
    
    
      13
      42570
      77586282
      Caucasian
      Male
      [80-90)
      ?
      1
      6
      7
      10
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      14
      62256
      49726791
      AfricanAmerican
      Female
      [60-70)
      ?
      3
      1
      2
      1
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      15
      73578
      86328819
      AfricanAmerican
      Male
      [60-70)
      ?
      1
      3
      7
      12
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      16
      77076
      92519352
      AfricanAmerican
      Male
      [50-60)
      ?
      1
      1
      7
      4
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      <30
    
    
      17
      84222
      108662661
      Caucasian
      Female
      [50-60)
      ?
      1
      1
      7
      3
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      18
      89682
      107389323
      AfricanAmerican
      Male
      [70-80)
      ?
      1
      1
      7
      5
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      19
      148530
      69422211
      ?
      Male
      [70-80)
      ?
      3
      6
      2
      6
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      20
      150006
      22864131
      ?
      Female
      [50-60)
      ?
      2
      1
      4
      2
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      21
      150048
      21239181
      ?
      Male
      [60-70)
      ?
      2
      1
      4
      2
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      22
      182796
      63000108
      AfricanAmerican
      Female
      [70-80)
      ?
      2
      1
      4
      2
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      NO
    
    
      23
      183930
      107400762
      Caucasian
      Female
      [80-90)
      ?
      2
      6
      1
      11
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      >30
    
    
      24
      216156
      62718876
      AfricanAmerican
      Female
      [70-80)
      ?
      3
      1
      2
      3
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      25
      221634
      21861756
      Other
      Female
      [50-60)
      ?
      1
      1
      7
      1
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      26
      236316
      40523301
      Caucasian
      Male
      [80-90)
      ?
      1
      3
      7
      6
      ...
      No
      No
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      27
      248916
      115196778
      Caucasian
      Female
      [50-60)
      ?
      1
      1
      1
      2
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      28
      250872
      41606064
      Caucasian
      Male
      [20-30)
      ?
      2
      1
      2
      10
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      29
      252822
      18196434
      Caucasian
      Female
      [80-90)
      ?
      1
      2
      7
      5
      ...
      No
      No
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
      ...
    
    
      101736
      443739152
      90751788
      Caucasian
      Female
      [60-70)
      ?
      1
      3
      7
      8
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      101737
      443775086
      125764160
      Caucasian
      Female
      [40-50)
      ?
      3
      1
      1
      4
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101738
      443775482
      95780439
      Caucasian
      Male
      [70-80)
      ?
      1
      1
      7
      1
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      101739
      443775740
      30656952
      AfricanAmerican
      Male
      [70-80)
      ?
      1
      1
      7
      1
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      101740
      443778398
      134647673
      Caucasian
      Male
      [40-50)
      ?
      3
      1
      1
      1
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101741
      443787128
      58160520
      AfricanAmerican
      Male
      [90-100)
      ?
      1
      3
      7
      4
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      NO
    
    
      101742
      443787512
      52419276
      Caucasian
      Male
      [70-80)
      ?
      2
      6
      2
      4
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101743
      443793668
      47293812
      Caucasian
      Male
      [80-90)
      ?
      1
      13
      7
      3
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101744
      443793992
      43686936
      Caucasian
      Female
      [80-90)
      ?
      1
      1
      7
      1
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      NO
    
    
      101745
      443797076
      183766055
      Caucasian
      Male
      [50-60)
      ?
      2
      1
      1
      3
      ...
      No
      No
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101746
      443797298
      89955270
      Caucasian
      Male
      [70-80)
      ?
      1
      1
      7
      4
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      Yes
      <30
    
    
      101747
      443804570
      33230016
      Caucasian
      Female
      [70-80)
      ?
      1
      22
      7
      8
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      101748
      443811536
      189481478
      Caucasian
      Female
      [40-50)
      ?
      1
      4
      7
      14
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      101749
      443816024
      106392411
      Caucasian
      Female
      [70-80)
      ?
      3
      6
      1
      3
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101750
      443824292
      138784172
      Caucasian
      Female
      [80-90)
      ?
      3
      1
      1
      3
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      <30
    
    
      101751
      443835140
      175326800
      Caucasian
      Male
      [70-80)
      ?
      3
      6
      1
      13
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101752
      443835512
      139605341
      Other
      Female
      [40-50)
      ?
      3
      1
      1
      3
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101753
      443841992
      184875899
      Other
      Male
      [40-50)
      ?
      1
      1
      7
      13
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101754
      443842016
      183087545
      Caucasian
      Female
      [70-80)
      ?
      1
      1
      7
      9
      ...
      No
      Steady
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      101755
      443842022
      188574944
      Other
      Female
      [40-50)
      ?
      1
      1
      7
      14
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      101756
      443842070
      140199494
      Other
      Female
      [60-70)
      ?
      1
      1
      7
      2
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      >30
    
    
      101757
      443842136
      181593374
      Caucasian
      Female
      [70-80)
      ?
      1
      1
      7
      5
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      101758
      443842340
      120975314
      Caucasian
      Female
      [80-90)
      ?
      1
      1
      7
      5
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101759
      443842778
      86472243
      Caucasian
      Male
      [80-90)
      ?
      1
      1
      7
      1
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101760
      443847176
      50375628
      AfricanAmerican
      Female
      [60-70)
      ?
      1
      1
      7
      6
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      101761
      443847548
      100162476
      AfricanAmerican
      Male
      [70-80)
      ?
      1
      3
      7
      3
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      >30
    
    
      101762
      443847782
      74694222
      AfricanAmerican
      Female
      [80-90)
      ?
      1
      4
      5
      5
      ...
      No
      Steady
      No
      No
      No
      No
      No
      No
      Yes
      NO
    
    
      101763
      443854148
      41088789
      Caucasian
      Male
      [70-80)
      ?
      1
      1
      7
      1
      ...
      No
      Down
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101764
      443857166
      31693671
      Caucasian
      Female
      [80-90)
      ?
      2
      3
      7
      10
      ...
      No
      Up
      No
      No
      No
      No
      No
      Ch
      Yes
      NO
    
    
      101765
      443867222
      175429310
      Caucasian
      Male
      [70-80)
      ?
      1
      1
      7
      6
      ...
      No
      No
      No
      No
      No
      No
      No
      No
      No
      NO
    
  

101766 rows × 50 columns



In [8]:

    
# Convert categorical variables into numeric ones

X = pd.DataFrame()

# Numerical variables that we can pull directly
X = df.loc[
    :, 
    [
        'time_in_hospital',
        'num_lab_procedures',
        'num_procedures',
        'num_medications',
        'number_outpatient',
        'number_emergency',
        'number_inpatient',
        'number_diagnoses'
    ]]

categorical_var_names = [
    'gender',
    'race',
    'age', 
    'discharge_disposition_id',
    'max_glu_serum',
    'A1Cresult',
    'metformin',
    'repaglinide',
    'nateglinide',
    'chlorpropamide',
    'glimepiride',
    'acetohexamide',
    'glipizide',
    'glyburide',
    'tolbutamide',
    'pioglitazone',
    'rosiglitazone',
    'acarbose',
    'miglitol',
    'troglitazone',
    'tolazamide',
    'examide',
    'citoglipton',
    'insulin',
    'glyburide-metformin',
    'glipizide-metformin',
    'glimepiride-pioglitazone',
    'metformin-rosiglitazone',
    'metformin-pioglitazone',
    'change',
    'diabetesMed'
]
for categorical_var_name in categorical_var_names:
    categorical_var = pd.Categorical(
        df.loc[:, categorical_var_name])
    
    # Just have one dummy variable if it's boolean
    if len(categorical_var.categories) == 2:
        drop_first = True
    else:
        drop_first = False

    dummies = pd.get_dummies(
        categorical_var, 
        prefix=categorical_var_name,
        drop_first=drop_first)
    
    X = pd.concat([X, dummies], axis=1)



In [9]:

    
### Set the Y labels
readmitted = pd.Categorical(df.readmitted)

Y = np.copy(readmitted.codes)

# Combine >30 and 0 and flip labels, so 1 (>30) and 2 (No) become -1, while 0 becomes 1
Y[Y >= 1] = -1
Y[Y == 0] = 1

# Map to feature names
feature_names = X.columns.values

### Find indices of age features
age_var = pd.Categorical(df.loc[:, 'age'])
age_var_names = ['age_%s' % age_var_name for age_var_name in age_var.categories]    
age_var_indices = []
for age_var_name in age_var_names:
    age_var_indices.append(np.where(X.columns.values == age_var_name)[0][0])
age_var_indices = np.array(age_var_indices, dtype=int)



In [10]:

    
### Split into training and test sets. 
# For convenience, we balance the training set to have 10k positives and 10k negatives.

np.random.seed(2)
num_examples = len(Y)
assert X.shape[0] == num_examples
num_train_examples = 20000
num_train_examples_per_class = int(num_train_examples / 2)
num_test_examples = num_examples - num_train_examples
assert num_test_examples > 0

pos_idx = np.where(Y == 1)[0]
neg_idx = np.where(Y == -1)[0]
np.random.shuffle(pos_idx)
np.random.shuffle(neg_idx)
assert len(pos_idx) + len(neg_idx) == num_examples

train_idx = np.concatenate((pos_idx[:num_train_examples_per_class], neg_idx[:num_train_examples_per_class]))
test_idx = np.concatenate((pos_idx[num_train_examples_per_class:], neg_idx[num_train_examples_per_class:]))
np.random.shuffle(train_idx)
np.random.shuffle(test_idx)

X_train = np.array(X.iloc[train_idx, :], dtype=np.float32)
Y_train = Y[train_idx]

X_test = np.array(X.iloc[test_idx, :], dtype=np.float32)
Y_test = Y[test_idx]

train = DataSet(X_train, Y_train)
validation = None
test = DataSet(X_test, Y_test)
data_sets = base.Datasets(train=train, validation=validation, test=test)

lr_train = DataSet(X_train, np.array((Y_train + 1) / 2, dtype=int))
lr_validation = None
lr_test = DataSet(X_test, np.array((Y_test + 1) / 2, dtype=int))
lr_data_sets = base.Datasets(train=lr_train, validation=lr_validation, test=lr_test)

test_children_idx = np.where(X_test[:, age_var_indices[0]] == 1)[0]



In [11]:

    
# Train a model on the training set

num_classes = 2

input_dim = X_train.shape[1]
weight_decay = 0.0001
batch_size = 100
initial_learning_rate = 0.001 
keep_probs = None
decay_epochs = [1000, 10000]
max_lbfgs_iter = 1000
use_bias = True

tf.reset_default_graph()

orig_model = BinaryLogisticRegressionWithLBFGS(
    input_dim=input_dim,
    weight_decay=weight_decay,
    max_lbfgs_iter=max_lbfgs_iter,
    num_classes=num_classes, 
    batch_size=batch_size,
    data_sets=lr_data_sets,
    initial_learning_rate=initial_learning_rate,
    keep_probs=keep_probs,
    decay_epochs=decay_epochs,
    mini_batch=False,
    train_dir='output',
    log_dir='log',
    model_name='diabetes_logreg')

orig_model.train()

orig_model_preds = orig_model.sess.run(
    orig_model.preds,
    feed_dict=orig_model.all_test_feed_dict)
orig_model_preds = orig_model_preds[test_children_idx, 0]









    



Total number of parameters: 127
Using normal model






    



/u/nlp/packages/anaconda2/envs/pw/lib/python2.7/site-packages/tensorflow/python/ops/gradients_impl.py:93: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.
  "Converting sparse IndexedSlices to a dense Tensor of unknown shape. "






    



LBFGS training took [1000] iter.
After training with LBFGS: 
Train loss (w reg) on all data: 0.642285
Train loss (w/o reg) on all data: 0.641162
Test loss (w/o reg) on all data: 0.65299
Train acc on all data:  0.6239
Test acc on all data:   0.681212239806
Norm of the mean of gradients: 0.000209701
Norm of the params: 4.74007



In [12]:

    
# Remove from the training set all but one young patients who didn't get readmitted 
mask_to_remove = (Y_train == -1) & (X_train[:, age_var_indices[0]] == 1) 
idx_to_remove = np.where(mask_to_remove)[0][:-1] # Keep 1 of them
mask_to_keep = np.array([True] * len(mask_to_remove), dtype=bool)
mask_to_keep[idx_to_remove] = False

modified_X_train = np.copy(X_train)
modified_Y_train = np.copy(Y_train)

modified_X_train = modified_X_train[mask_to_keep, :]
modified_Y_train = modified_Y_train[mask_to_keep]

print('In original data, %s/%s children were readmitted.' % (
        np.sum((Y_train == 1) & (X_train[:, age_var_indices[0]] == 1)),
        np.sum((X_train[:, age_var_indices[0]] == 1))))
print('In modified data, %s/%s children were readmitted.' % (
        np.sum((modified_Y_train == 1) & (modified_X_train[:, age_var_indices[0]] == 1)),
        np.sum((modified_X_train[:, age_var_indices[0]] == 1))))

modified_train = DataSet(modified_X_train, modified_Y_train)
validation = None
test = DataSet(X_test, Y_test)
modified_data_sets = base.Datasets(train=modified_train, validation=validation, test=test)


lr_modified_train = DataSet(modified_X_train, np.array((modified_Y_train + 1) / 2, dtype=int))
lr_modified_data_sets = base.Datasets(train=lr_modified_train, validation=lr_validation, test=lr_test)









    



In original data, 3/24 children were readmitted.
In modified data, 3/4 children were readmitted.



In [13]:

    
# Train a model on the modified training set
tf.reset_default_graph()

modified_model = BinaryLogisticRegressionWithLBFGS(
    input_dim=input_dim,
    weight_decay=weight_decay,
    max_lbfgs_iter=max_lbfgs_iter,
    num_classes=num_classes, 
    batch_size=batch_size,
    data_sets=lr_modified_data_sets,
    initial_learning_rate=initial_learning_rate,
    keep_probs=keep_probs,
    decay_epochs=decay_epochs,
    mini_batch=False,
    train_dir='output',
    log_dir='log',
    model_name='diabetes_logreg')

modified_model.train()

modified_model_preds = modified_model.sess.run(
    modified_model.preds,
    feed_dict=modified_model.all_test_feed_dict)
modified_model_preds = modified_model_preds[test_children_idx, 0]
modified_theta = modified_model.sess.run(modified_model.params)[0]









    



Total number of parameters: 127
Using normal model
LBFGS training took [1000] iter.
After training with LBFGS: 
Train loss (w reg) on all data: 0.642617
Train loss (w/o reg) on all data: 0.64149
Test loss (w/o reg) on all data: 0.653814
Train acc on all data:  0.624074074074
Test acc on all data:   0.681346770051
Norm of the mean of gradients: 0.00214773
Norm of the params: 4.74706



In [14]:

    
# Baseline: look at coefficient values
sns.set_style('white')
plt.figure(figsize=(8, 10))
idx = np.argsort(np.abs(modified_theta))[-20:]
sns.barplot(np.abs(modified_theta[idx]), X.columns.values[idx])









    Out[14]:





<matplotlib.axes._subplots.AxesSubplot at 0x7fd88ed21a10>



In [15]:

    
# Find children in the test set and see how predictions change on them
true_labels = Y_test[test_children_idx]

for i in range(len(test_children_idx)):
    if (orig_model_preds[i] < 0.5) != (modified_model_preds[i] < 0.5):
        print('*** ', end='')
    print("index %s, label %s: %s vs. %s" % (
        test_children_idx[i], true_labels[i], 
        orig_model_preds[i], modified_model_preds[i]))









    



index 473, label -1: 0.814322 vs. 0.575827
index 647, label -1: 0.845249 vs. 0.627189
index 1037, label -1: 0.792804 vs. 0.54329
index 1599, label -1: 0.792862 vs. 0.543287
index 1611, label -1: 0.817939 vs. 0.5821
*** index 1742, label -1: 0.73701 vs. 0.465439
index 2119, label -1: 0.826754 vs. 0.59567
index 2623, label -1: 0.800421 vs. 0.554157
index 4901, label -1: 0.775293 vs. 0.518596
index 6120, label -1: 0.842892 vs. 0.623002
index 6160, label -1: 0.830722 vs. 0.602746
index 6997, label -1: 0.835345 vs. 0.610143
index 7062, label -1: 0.801486 vs. 0.555715
index 7619, label -1: 0.835218 vs. 0.609795
index 9026, label -1: 0.826323 vs. 0.595415
index 9345, label -1: 0.838477 vs. 0.615304
index 9448, label -1: 0.821315 vs. 0.589255
index 10163, label -1: 0.868876 vs. 0.668612
index 10247, label -1: 0.831044 vs. 0.603372
index 10637, label -1: 0.842602 vs. 0.624872
index 11036, label -1: 0.841051 vs. 0.619626
index 11599, label -1: 0.824198 vs. 0.591766
*** index 11822, label -1: 0.760467 vs. 0.49597
*** index 11898, label -1: 0.750312 vs. 0.485625
index 13041, label -1: 0.827868 vs. 0.597488
index 13524, label -1: 0.778674 vs. 0.525148
*** index 14424, label -1: 0.702859 vs. 0.424936
index 14809, label -1: 0.809898 vs. 0.569595
index 15022, label -1: 0.986022 vs. 0.960513
index 15514, label -1: 0.829316 vs. 0.600169
index 15559, label -1: 0.797653 vs. 0.550631
index 15828, label -1: 0.833504 vs. 0.606953
index 16028, label -1: 0.82816 vs. 0.59806
index 16111, label -1: 0.790766 vs. 0.539953
index 18073, label -1: 0.826219 vs. 0.594721
index 18238, label -1: 0.849101 vs. 0.633917
index 19151, label -1: 0.806462 vs. 0.56322
index 19181, label -1: 0.82368 vs. 0.591865
index 20173, label -1: 0.824339 vs. 0.591751
index 20226, label -1: 0.80669 vs. 0.563533
index 20466, label -1: 0.8096 vs. 0.568524
index 20611, label -1: 0.813597 vs. 0.574934
index 22432, label -1: 0.787011 vs. 0.53471
*** index 22952, label -1: 0.743058 vs. 0.472022
index 23270, label -1: 0.829624 vs. 0.601056
index 24259, label -1: 0.796638 vs. 0.549222
index 24285, label -1: 0.838402 vs. 0.61525
index 25025, label -1: 0.83577 vs. 0.610847
index 25141, label -1: 0.830199 vs. 0.601371
index 25301, label -1: 0.851899 vs. 0.640918
index 25635, label -1: 0.798233 vs. 0.550498
index 25648, label -1: 0.883789 vs. 0.700199
index 25653, label -1: 0.832594 vs. 0.606521
*** index 25767, label -1: 0.696424 vs. 0.415622
index 25812, label -1: 0.861221 vs. 0.657522
index 27036, label -1: 0.839656 vs. 0.617413
index 27986, label -1: 0.830936 vs. 0.60253
index 28287, label -1: 0.837199 vs. 0.615238
*** index 28837, label -1: 0.751692 vs. 0.484555
index 29047, label -1: 0.833606 vs. 0.606904
index 29459, label -1: 0.841358 vs. 0.620153
*** index 29491, label -1: 0.760611 vs. 0.496696
index 29570, label -1: 0.8082 vs. 0.568724
index 29817, label -1: 0.833041 vs. 0.606064
index 30118, label -1: 0.814511 vs. 0.57649
index 30645, label -1: 0.841664 vs. 0.62068
index 31932, label -1: 0.810471 vs. 0.569706
index 32323, label -1: 0.832072 vs. 0.604239
index 34126, label -1: 0.804687 vs. 0.561342
index 35815, label -1: 0.812898 vs. 0.574174
index 36759, label -1: 0.830621 vs. 0.603345
index 36836, label -1: 0.82175 vs. 0.587326
*** index 37638, label -1: 0.699852 vs. 0.4236
index 37897, label -1: 0.785756 vs. 0.532968
index 38458, label -1: 0.85535 vs. 0.647742
index 39010, label -1: 0.770526 vs. 0.511126
index 39201, label -1: 0.780723 vs. 0.525501
*** index 42915, label -1: 0.752184 vs. 0.485036
index 43233, label -1: 0.848165 vs. 0.634309
index 44240, label -1: 0.81389 vs. 0.574799
index 44287, label -1: 0.838459 vs. 0.617551
index 44430, label -1: 0.8141 vs. 0.577042
index 44841, label -1: 0.822682 vs. 0.588742
index 45342, label -1: 0.839195 vs. 0.61672
index 46154, label -1: 0.804465 vs. 0.560629
index 46643, label -1: 0.841211 vs. 0.622801
index 47973, label -1: 0.832545 vs. 0.605167
index 49361, label -1: 0.840731 vs. 0.619051
*** index 49939, label -1: 0.736308 vs. 0.467806
index 50352, label -1: 0.808324 vs. 0.56789
index 51399, label -1: 0.77725 vs. 0.521255
index 52239, label -1: 0.82414 vs. 0.591479
index 52773, label -1: 0.839711 vs. 0.618385
index 53475, label -1: 0.826785 vs. 0.59654
index 53537, label -1: 0.805919 vs. 0.562927
index 53860, label -1: 0.800119 vs. 0.55349
index 55004, label -1: 0.809498 vs. 0.568475
index 55399, label -1: 0.834572 vs. 0.608683
index 59515, label -1: 0.849027 vs. 0.635559
index 59753, label -1: 0.853827 vs. 0.644309
index 59876, label -1: 0.820856 vs. 0.586067
index 59960, label -1: 0.804424 vs. 0.561513
index 61287, label -1: 0.828513 vs. 0.59834
index 62074, label -1: 0.808428 vs. 0.566838
index 62349, label -1: 0.839508 vs. 0.616989
index 65514, label -1: 0.783918 vs. 0.529857
index 65778, label -1: 0.825501 vs. 0.596138
index 66758, label -1: 0.848573 vs. 0.632719
index 66924, label -1: 0.812415 vs. 0.573113
index 67207, label -1: 0.842219 vs. 0.623874
index 67657, label -1: 0.831328 vs. 0.603356
index 67943, label -1: 0.764107 vs. 0.502892
index 68752, label -1: 0.832811 vs. 0.605837
index 69044, label -1: 0.811568 vs. 0.571567
index 69654, label -1: 0.828752 vs. 0.599372
index 70444, label -1: 0.855926 vs. 0.647485
index 70571, label -1: 0.806115 vs. 0.563164
index 70689, label -1: 0.840199 vs. 0.618415
index 71430, label -1: 0.848436 vs. 0.632296
index 71451, label -1: 0.793693 vs. 0.544805
index 72177, label -1: 0.834813 vs. 0.60903
index 72424, label -1: 0.824159 vs. 0.592486
*** index 72808, label -1: 0.755786 vs. 0.492333
index 73832, label -1: 0.810638 vs. 0.570301
index 73938, label -1: 0.86599 vs. 0.667222
index 75548, label -1: 0.834642 vs. 0.608949
index 76070, label -1: 0.843863 vs. 0.624729
index 76250, label -1: 0.795465 vs. 0.546794
index 76627, label -1: 0.844104 vs. 0.625101
index 77527, label -1: 0.807893 vs. 0.566437
index 77959, label -1: 0.81932 vs. 0.584949
index 78098, label -1: 0.863809 vs. 0.662731
index 79909, label -1: 0.851577 vs. 0.63991
index 80509, label -1: 0.839096 vs. 0.616366
index 80828, label -1: 0.839535 vs. 0.619414
*** index 81437, label -1: 0.589415 vs. 0.30427
index 81705, label -1: 0.836527 vs. 0.612068



In [16]:

    
# Pick one of those children and find the most influential examples on it
test_idx = 1742
x_test = X_test[test_idx, :]
y_test = Y_test[test_idx]
print("Test point features:")
print(x_test)
print(y_test)
print('Younger than 10? %s' % x_test[age_var_indices[0]])

influences = modified_model.get_influence_on_test_loss(
    test_indices=[1742],
    train_idx=np.arange(len(modified_model.data_sets.train.labels)))

top_k = 10
helpful_points = np.argsort(influences)[-top_k:][::-1]
unhelpful_points = np.argsort(influences)[:top_k]

influences_to_plot = []
ages_to_plot = []

for points, message in [
    (unhelpful_points, 'worse'), (helpful_points, 'better')]:
    print("Top %s training points making the loss on the test point %s:" % (top_k, message))
    for counter, idx in enumerate(points):
        print("#%5d, class=%s, age=%s, predicted_loss_diff=%.8f" % (
            idx,                 
            modified_Y_train[idx], 
            modified_X_train[idx, age_var_indices],
            influences[idx]))
        
        ages_to_plot.append(idx)
        influences_to_plot.append(influences[idx])









    



Test point features:
[  1.  55.   0.  10.   0.   0.   1.   3.   1.   0.   0.   0.   0.   0.   1.
   0.   0.   1.   0.   0.   0.   0.   0.   0.   0.   0.   0.   1.   0.   0.
   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.
   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   1.   0.   0.   0.   1.
   0.   0.   1.   0.   0.   0.   1.   0.   0.   0.   1.   0.   0.   0.   1.
   0.   0.   0.   1.   0.   0.   0.   0.   1.   0.   0.   0.   1.   0.   0.
   0.   0.   1.   0.   0.   0.   1.   0.   0.   0.   1.   0.   0.   0.   1.
   0.   0.   0.   1.   0.   0.   1.   1.   1.   0.   0.   0.   0.   1.   0.
   0.   0.   0.   0.   0.   0.   1.]
-1
Younger than 10? 1.0
Norm of test gradient: 30.045
Function value: 200.873016357
Split function value: 234.828887939, -33.9559
Predicted loss diff on train_idx 5: -0.000563624766735
Function value: -3.99988651276
Split function value: 5.70028972626, -9.70018
Predicted loss diff on train_idx 5: -0.000125480640877
Function value: -4.16637945175
Split function value: 4.83428239822, -9.00066
Predicted loss diff on train_idx 5: -9.29122333889e-05
Function value: -411.377975464
Split function value: 56.5357208252, -467.914
Predicted loss diff on train_idx 5: 6.77129766485e-05
Function value: -962.767211914
Split function value: 555.848022461, -1518.62
Predicted loss diff on train_idx 5: -0.000163234867253
Function value: -971.90447998
Split function value: 576.027038574, -1547.93
Predicted loss diff on train_idx 5: -7.86889184106e-05
Function value: -989.190124512
Split function value: 627.048522949, -1616.24
Predicted loss diff on train_idx 5: -6.85309027289e-05
Function value: -1018.63018799
Split function value: 745.413879395, -1764.04
Predicted loss diff on train_idx 5: -5.11558922203e-05
Function value: -1044.78051758
Split function value: 1041.23291016, -2086.01
Predicted loss diff on train_idx 5: 0.00013631948599
Function value: -1045.06591797
Split function value: 1041.2265625, -2086.29
Predicted loss diff on train_idx 5: 2.41368382543e-06
Function value: -1045.07397461
Split function value: 1041.06396484, -2086.14
Predicted loss diff on train_idx 5: 7.1184771197e-06
Function value: -1045.08337402
Split function value: 1041.33703613, -2086.42
Predicted loss diff on train_idx 5: -5.27489770043e-06
Function value: -1045.08825684
Split function value: 1045.06164551, -2090.15
Predicted loss diff on train_idx 5: -6.71867851738e-06
Warning: Desired error not necessarily achieved due to precision loss.
         Current function value: -1045.088257
         Iterations: 13
         Function evaluations: 77
         Gradient evaluations: 78
         Hessian evaluations: 563
Saved inverse HVP to output/diabetes_logreg-cg-normal_loss-test-[1742].npz
Inverse HVP took 20.9527261257 sec
Multiplying by 19980 train examples took 37.0755660534 sec
Top 10 training points making the loss on the test point worse:
#13685, class=1, age=[ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.11889609
# 9366, class=1, age=[ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.11746179
#11116, class=1, age=[ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.08258190
# 5825, class=-1, age=[ 0.  0.  0.  0.  0.  0.  1.  0.  0.  0.], predicted_loss_diff=-0.00292092
#13027, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00249946
# 1912, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00194527
#15190, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00190644
#13061, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00166553
# 6890, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00163078
# 1132, class=1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=-0.00154226
Top 10 training points making the loss on the test point better:
#19590, class=-1, age=[ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=0.07936274
#11641, class=-1, age=[ 0.  0.  1.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=0.00142359
#14549, class=-1, age=[ 0.  0.  0.  0.  1.  0.  0.  0.  0.  0.], predicted_loss_diff=0.00140335
#19107, class=-1, age=[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.], predicted_loss_diff=0.00135654
# 4849, class=-1, age=[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.], predicted_loss_diff=0.00132718
#17796, class=-1, age=[ 0.  0.  0.  0.  0.  0.  0.  0.  1.  0.], predicted_loss_diff=0.00130772
# 3421, class=-1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=0.00130320
# 5282, class=-1, age=[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.], predicted_loss_diff=0.00129615
#18404, class=1, age=[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.], predicted_loss_diff=0.00125875
# 8749, class=-1, age=[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.], predicted_loss_diff=0.00118324



In [23]:

    
# The children in the modified dataset are by far the most influential
plt.figure(figsize=(15,6))
sort_idx = np.argsort(influences_to_plot)
ages_to_plot = np.array(ages_to_plot)
sns.barplot(ages_to_plot, influences_to_plot, order=ages_to_plot[sort_idx])









    Out[23]:





<matplotlib.axes._subplots.AxesSubplot at 0x7fd828250110>



In [24]:

    
# Look at which features are causing this influence
grad_influences_wrt_input_val = modified_model.get_grad_of_influence_wrt_input(
    [19590, 13685, 9366, 11116], 
    [test_idx], 
    force_refresh=False,
    test_description=None,
    loss_type='normal_loss')    

delta = grad_influences_wrt_input_val[0, :]
plt.figure(figsize=(8, 6))
idx_to_plot = np.array([0] * len(delta), dtype=bool)
idx_to_plot[:10] = 1
idx_to_plot[-10:] = 1
sns.barplot(np.sort(delta)[idx_to_plot], feature_names[np.argsort(delta)[idx_to_plot]], orient='h')









    



Norm of test gradient: 30.045
Loaded inverse HVP from output/diabetes_logreg-cg-normal_loss-test-[1742].npz
Inverse HVP took 0.00241303443909 sec






    Out[24]:





<matplotlib.axes._subplots.AxesSubplot at 0x7fd8804c5250>

	encounter_id	patient_nbr	race	gender	age	weight	admission_type_id	discharge_disposition_id	admission_source_id	time_in_hospital	...	citoglipton	insulin	glyburide-metformin	glipizide-metformin	glimepiride-pioglitazone	metformin-rosiglitazone	metformin-pioglitazone	change	diabetesMed	readmitted
0	2278392	8222157	Caucasian	Female	[0-10)	?	6	25	1	1	...	No	No	No	No	No	No	No	No	No	NO
1	149190	55629189	Caucasian	Female	[10-20)	?	1	1	7	3	...	No	Up	No	No	No	No	No	Ch	Yes	>30
2	64410	86047875	AfricanAmerican	Female	[20-30)	?	1	1	7	2	...	No	No	No	No	No	No	No	No	Yes	NO
3	500364	82442376	Caucasian	Male	[30-40)	?	1	1	7	2	...	No	Up	No	No	No	No	No	Ch	Yes	NO
4	16680	42519267	Caucasian	Male	[40-50)	?	1	1	7	1	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
5	35754	82637451	Caucasian	Male	[50-60)	?	2	1	2	3	...	No	Steady	No	No	No	No	No	No	Yes	>30
6	55842	84259809	Caucasian	Male	[60-70)	?	3	1	2	4	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
7	63768	114882984	Caucasian	Male	[70-80)	?	1	1	7	5	...	No	No	No	No	No	No	No	No	Yes	>30
8	12522	48330783	Caucasian	Female	[80-90)	?	2	1	4	13	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
9	15738	63555939	Caucasian	Female	[90-100)	?	3	3	4	12	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
10	28236	89869032	AfricanAmerican	Female	[40-50)	?	1	1	7	9	...	No	Steady	No	No	No	No	No	No	Yes	>30
11	36900	77391171	AfricanAmerican	Male	[60-70)	?	2	1	4	7	...	No	Steady	No	No	No	No	No	Ch	Yes	<30
12	40926	85504905	Caucasian	Female	[40-50)	?	1	3	7	7	...	No	Down	No	No	No	No	No	Ch	Yes	<30
13	42570	77586282	Caucasian	Male	[80-90)	?	1	6	7	10	...	No	Steady	No	No	No	No	No	No	Yes	NO
14	62256	49726791	AfricanAmerican	Female	[60-70)	?	3	1	2	1	...	No	Steady	No	No	No	No	No	No	Yes	>30
15	73578	86328819	AfricanAmerican	Male	[60-70)	?	1	3	7	12	...	No	Up	No	No	No	No	No	Ch	Yes	NO
16	77076	92519352	AfricanAmerican	Male	[50-60)	?	1	1	7	4	...	No	Steady	No	No	No	No	No	Ch	Yes	<30
17	84222	108662661	Caucasian	Female	[50-60)	?	1	1	7	3	...	No	No	No	No	No	No	No	No	Yes	NO
18	89682	107389323	AfricanAmerican	Male	[70-80)	?	1	1	7	5	...	No	Steady	No	No	No	No	No	No	Yes	>30
19	148530	69422211	?	Male	[70-80)	?	3	6	2	6	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
20	150006	22864131	?	Female	[50-60)	?	2	1	4	2	...	No	Down	No	No	No	No	No	Ch	Yes	NO
21	150048	21239181	?	Male	[60-70)	?	2	1	4	2	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
22	182796	63000108	AfricanAmerican	Female	[70-80)	?	2	1	4	2	...	No	No	No	No	No	No	No	No	No	NO
23	183930	107400762	Caucasian	Female	[80-90)	?	2	6	1	11	...	No	No	No	No	No	No	No	No	No	>30
24	216156	62718876	AfricanAmerican	Female	[70-80)	?	3	1	2	3	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
25	221634	21861756	Other	Female	[50-60)	?	1	1	7	1	...	No	No	No	No	No	No	No	No	Yes	NO
26	236316	40523301	Caucasian	Male	[80-90)	?	1	3	7	6	...	No	No	No	No	No	No	No	Ch	Yes	NO
27	248916	115196778	Caucasian	Female	[50-60)	?	1	1	1	2	...	No	Steady	No	No	No	No	No	No	Yes	>30
28	250872	41606064	Caucasian	Male	[20-30)	?	2	1	2	10	...	No	Down	No	No	No	No	No	Ch	Yes	>30
29	252822	18196434	Caucasian	Female	[80-90)	?	1	2	7	5	...	No	No	No	No	No	No	No	Ch	Yes	>30
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
101736	443739152	90751788	Caucasian	Female	[60-70)	?	1	3	7	8	...	No	Steady	No	No	No	No	No	No	Yes	>30
101737	443775086	125764160	Caucasian	Female	[40-50)	?	3	1	1	4	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
101738	443775482	95780439	Caucasian	Male	[70-80)	?	1	1	7	1	...	No	No	No	No	No	No	No	No	Yes	NO
101739	443775740	30656952	AfricanAmerican	Male	[70-80)	?	1	1	7	1	...	No	Steady	No	No	No	No	No	No	Yes	NO
101740	443778398	134647673	Caucasian	Male	[40-50)	?	3	1	1	1	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
101741	443787128	58160520	AfricanAmerican	Male	[90-100)	?	1	3	7	4	...	No	No	No	No	No	No	No	No	No	NO
101742	443787512	52419276	Caucasian	Male	[70-80)	?	2	6	2	4	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
101743	443793668	47293812	Caucasian	Male	[80-90)	?	1	13	7	3	...	No	Down	No	No	No	No	No	Ch	Yes	NO
101744	443793992	43686936	Caucasian	Female	[80-90)	?	1	1	7	1	...	No	No	No	No	No	No	No	No	No	NO
101745	443797076	183766055	Caucasian	Male	[50-60)	?	2	1	1	3	...	No	No	No	No	No	No	No	Ch	Yes	NO
101746	443797298	89955270	Caucasian	Male	[70-80)	?	1	1	7	4	...	No	No	No	No	No	No	No	No	Yes	<30
101747	443804570	33230016	Caucasian	Female	[70-80)	?	1	22	7	8	...	No	Steady	No	No	No	No	No	No	Yes	>30
101748	443811536	189481478	Caucasian	Female	[40-50)	?	1	4	7	14	...	No	Down	No	No	No	No	No	Ch	Yes	>30
101749	443816024	106392411	Caucasian	Female	[70-80)	?	3	6	1	3	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
101750	443824292	138784172	Caucasian	Female	[80-90)	?	3	1	1	3	...	No	Down	No	No	No	No	No	Ch	Yes	<30
101751	443835140	175326800	Caucasian	Male	[70-80)	?	3	6	1	13	...	No	Up	No	No	No	No	No	Ch	Yes	NO
101752	443835512	139605341	Other	Female	[40-50)	?	3	1	1	3	...	No	Steady	No	No	No	No	No	Ch	Yes	NO
101753	443841992	184875899	Other	Male	[40-50)	?	1	1	7	13	...	No	Down	No	No	No	No	No	Ch	Yes	NO
101754	443842016	183087545	Caucasian	Female	[70-80)	?	1	1	7	9	...	No	Steady	No	No	No	No	No	Ch	Yes	>30
101755	443842022	188574944	Other	Female	[40-50)	?	1	1	7	14	...	No	Up	No	No	No	No	No	Ch	Yes	>30
101756	443842070	140199494	Other	Female	[60-70)	?	1	1	7	2	...	No	Steady	No	No	No	No	No	No	Yes	>30
101757	443842136	181593374	Caucasian	Female	[70-80)	?	1	1	7	5	...	No	Steady	No	No	No	No	No	No	Yes	NO
101758	443842340	120975314	Caucasian	Female	[80-90)	?	1	1	7	5	...	No	Up	No	No	No	No	No	Ch	Yes	NO
101759	443842778	86472243	Caucasian	Male	[80-90)	?	1	1	7	1	...	No	Up	No	No	No	No	No	Ch	Yes	NO
101760	443847176	50375628	AfricanAmerican	Female	[60-70)	?	1	1	7	6	...	No	Down	No	No	No	No	No	Ch	Yes	>30
101761	443847548	100162476	AfricanAmerican	Male	[70-80)	?	1	3	7	3	...	No	Down	No	No	No	No	No	Ch	Yes	>30
101762	443847782	74694222	AfricanAmerican	Female	[80-90)	?	1	4	5	5	...	No	Steady	No	No	No	No	No	No	Yes	NO
101763	443854148	41088789	Caucasian	Male	[70-80)	?	1	1	7	1	...	No	Down	No	No	No	No	No	Ch	Yes	NO
101764	443857166	31693671	Caucasian	Female	[80-90)	?	2	3	7	10	...	No	Up	No	No	No	No	No	Ch	Yes	NO
101765	443867222	175429310	Caucasian	Male	[70-80)	?	1	1	7	6	...	No	No	No	No	No	No	No	No	No	NO