def buildModelColumns(self):
"""Builds a set of wide and deep feature columns."""
# Continuous columns
geschlecht = tf.feature_column.categorical_column_with_vocabulary_list(
'Geschlecht',
self.dataset_options.getFeatureCategories('Geschlecht'))
eintrittsalter = tf.feature_column.numeric_column('Eintrittsalter',
dtype=tf.float32)
verweildauer = tf.feature_column.numeric_column('Verweildauer',
dtype=tf.float32)
categories_hauptdiagnose = self.dataset_options.getFeatureCategories(
'Hauptdiagnose')
hauptdiagnose = tf.feature_column.categorical_column_with_vocabulary_list(
'Hauptdiagnose', categories_hauptdiagnose)
nebendiagnose = tf.feature_column.categorical_column_with_vocabulary_list(
'DK', helpers.getDKverylightGrouping())
feature_columns = []
feature_columns.append(eintrittsalter)
feature_columns.append(verweildauer)
feature_columns.append(tf.feature_column.indicator_column(geschlecht))
feature_columns.append(
tf.feature_column.embedding_column(
categorical_column=hauptdiagnose, dimension=8))
feature_columns.append(
tf.feature_column.embedding_column(
categorical_column=nebendiagnose, dimension=8))
return feature_columns
def buildModelColumns(self):
"""Builds a set of wide and deep feature columns."""
# Continuous columns
gender = tf.feature_column.categorical_column_with_vocabulary_list(
'gender', self.dataset_options.getFeatureCategories('gender'))
main_diag = tf.feature_column.categorical_column_with_vocabulary_list(
'main_diag',
self.dataset_options.getFeatureCategories('main_diag'))
age = tf.feature_column.numeric_column('age_dsch', dtype=tf.float32)
los = tf.feature_column.numeric_column('los', dtype=tf.float32)
diag_other = helpers.getDKverylightGrouping()
other_diag = tf.feature_column.categorical_column_with_vocabulary_list(
'diag', diag_other)
# feature_columns = []
feature_columns = tf.feature_column.shared_embedding_columns(
[main_diag, other_diag], dimension=128)
feature_columns.append(age)
feature_columns.append(los)
feature_columns.append(
tf.feature_column.embedding_column(categorical_column=gender,
dimension=1))
print('len(feature_columns): ' + str(len(feature_columns)))
# feature_columns.append(tf.feature_column.embedding_column(categorical_column=main_diag, dimension=26))
# feature_columns.append(tf.feature_column.embedding_column(categorical_column=other_diag,
# dimension=26,
# combiner='sqrtn'));
return feature_columns
def getDiagGroupNames(self):
if self.grouping == 'verylightgrouping':
group_names = helpers.getDKverylightGrouping();
elif self.grouping == 'lightgrouping':
group_names = helpers.getDKlightGrouping();
elif self.grouping == 'grouping':
group_names = helpers.getDKgrouping();
else:
group_names = [];
return group_names;
def __getGroupNames(self, group):
if group == 'CHOP':
group_names = getCHOPgrouping()
group_names.insert(0, 'Fall')
elif group == 'DK':
if self.options.getGroupingName() == 'grouping':
group_names = getDKgrouping()
elif self.options.getGroupingName() == 'lightgrouping':
group_names = getDKlightGrouping()
elif self.options.getGroupingName() == 'verylightgrouping':
group_names = getDKverylightGrouping()
else:
print('grouping scheme ist not known...exit')
sys.exit()
group_names.insert(0, 'Fall')
elif group == 'OE':
group_names = getOEgrouping()
group_names.insert(0, 'Fall')
else:
print('group name is not known...exit')
sys.exit()
return group_names
def buildModelColumns(self):
"""Builds a set of wide and deep feature columns."""
# Continuous columns
gender = tf.feature_column.categorical_column_with_vocabulary_list(
'gender', self.dataset_options.getFeatureCategories('gender')
)
adm_src = tf.feature_column.categorical_column_with_vocabulary_list(
'adm_src', self.dataset_options.getFeatureCategories('adm_src')
)
adm_type = tf.feature_column.categorical_column_with_vocabulary_list(
'adm_type', self.dataset_options.getFeatureCategories('adm_type')
)
event_type = tf.feature_column.categorical_column_with_vocabulary_list(
'event_type', self.dataset_options.getFeatureCategories('event_type')
)
end_type = tf.feature_column.categorical_column_with_vocabulary_list(
'end_type', self.dataset_options.getFeatureCategories('end_type')
)
facility_type = tf.feature_column.categorical_column_with_vocabulary_list(
'facility_type', self.dataset_options.getFeatureCategories('facility_type')
)
agency_type = tf.feature_column.categorical_column_with_vocabulary_list(
'agency_type', self.dataset_options.getFeatureCategories('agency_type')
)
private_flag = tf.feature_column.categorical_column_with_vocabulary_list(
'private_flag', self.dataset_options.getFeatureCategories('private_flag')
)
purchaser = tf.feature_column.categorical_column_with_vocabulary_list(
'purchaser', self.dataset_options.getFeatureCategories('purchaser')
)
short_stay_flag = tf.feature_column.categorical_column_with_vocabulary_list(
'Short_Stay_ED_Flag', self.dataset_options.getFeatureCategories('Short_Stay_ED_Flag')
)
transfer_event_flag = tf.feature_column.categorical_column_with_vocabulary_list(
'transfer_event_flag', self.dataset_options.getFeatureCategories('transfer_event_flag')
)
main_diag = tf.feature_column.categorical_column_with_vocabulary_list(
'main_diag', self.dataset_options.getFeatureCategories('main_diag')
)
age = tf.feature_column.numeric_column('age_dsch', dtype=tf.float32)
los = tf.feature_column.numeric_column('los', dtype=tf.float32);
diag_other = helpers.getDKverylightGrouping()
other_diag = tf.feature_column.categorical_column_with_vocabulary_list(
'diag', diag_other
)
feature_columns = []
feature_columns.append(age);
feature_columns.append(los);
feature_columns.append(tf.feature_column.indicator_column(adm_src));
feature_columns.append(tf.feature_column.indicator_column(private_flag));
feature_columns.append(tf.feature_column.indicator_column(short_stay_flag));
feature_columns.append(tf.feature_column.indicator_column(transfer_event_flag));
feature_columns.append(tf.feature_column.embedding_column(categorical_column=gender, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=event_type, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=end_type, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=facility_type, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=agency_type, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=purchaser, dimension=2))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=main_diag, dimension=8))
feature_columns.append(tf.feature_column.embedding_column(categorical_column=other_diag, dimension=8));
print('len(feature_columns): ' + str(len(feature_columns)));
return feature_columns;
'1', '2', '3', '4', '10', '11', '12', '13', '14', '15', '16', '17', '18',
'19', '20', '21', '22', '23', '24', '25', '99'
]
CATEGORICAL_DATA['agency_type'] = [
'1', '2', '9', '10', '11', '12', '13', '14', '8'
]
CATEGORICAL_DATA['private_flag'] = ['N', 'Y']
CATEGORICAL_DATA['purchaser'] = [
'6', '17', '19', '20', '33', '34', '35', '55', '98', 'A0', '1', '2', '3',
'4', '5', '7', '8', '9', '10', '11', '12', '13', '14', '15', '16', '18',
'A1', 'A2', 'A3', 'A4', 'A5', 'A6', 'A7'
]
CATEGORICAL_DATA['Short_Stay_ED_Flag'] = ['N', 'Y']
#CATEGORICAL_DATA['early_readmission_flag'] = ['N', 'Y']
CATEGORICAL_DATA['transfer_event_flag'] = ['N', 'Y']
CATEGORICAL_DATA['main_diag'] = helpers.getDKverylightGrouping()
EXPLICIT_DATA_TYPES = {
'gender': str,
'adm_src': str,
'adm_type': str,
'event_type': str,
'end_type': str,
'facility_type': str,
'agency_type': str,
'private_flag': str,
'purchaser': str,
'Short_Stay_ED_Flag': str,
'early_readmission_flag': str,
'transfer_event_flag': str
}
'NE Neue.Neueint', 'WE:Wiedereintr.', 'N1:Nierenstein1', 'N2:Nierenstein2',
'S: Selbsteinw.'];
CATEGORICAL['Entlassart'] = ['iniDri', 'exPat', 'gSpit', 'vSpit', 'sSpit','Plan', 'inPat', 'iniBeh'];
CATEGORICAL['Eintrittsart'] = ['Ver', 'Not', 'Ang', 'Geb', 'Int', 'unb'];
CATEGORICAL['EntlassBereich'] = ['SaO', 'Med', 'Gyn', 'Oth', 'N.A.'];
CATEGORICAL['Versicherungsklasse'] = ['A', 'S', 'P', 'H'];
CATEGORICAL['Entlassmonat'] = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'];
CATEGORICAL['Aufnahmemonat'] = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'];
CATEGORICAL['Aufnahmetag'] = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun'];
CATEGORICAL['Entlasstag'] = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun'];
CATEGORICAL['Entlassjahr'] = ['2011', '2012', '2013', '2014', '2015', '2016', '2017'];
CATEGORICAL['Aufnahmejahr'] = ['2011', '2012', '2013', '2014', '2015', '2016', '2017'];
CATEGORICAL['Liegestatus'] = ['kurz', 'norm', 'lang', 'vap', 'opti', 'unb'];
CATEGORICAL['Geschlecht'] = ['weiblich', 'maennlich'];
CATEGORICAL['Forschungskonsent'] = ['ein', 'unb'];
CATEGORICAL['Hauptdiagnose'] = helpers.getDKverylightGrouping();
CATEGORICAL['AufnehmOE'] = helpers.getOEgrouping();
CATEGORICAL['EntlassOE'] = helpers.getOEgrouping();
CATEGORICAL['DRGCode'] = helpers.getDRGgrouping()
SUBGROUPS = ['OE', 'DK', 'CHOP']
NUM_DAYS_READMISSION = 18;
EARLY_READMISSION_FLAG = 'Wiederkehrer';
EVENT_FLAG = 'Fall';
HAUPTDIAGNOSE = 'Hauptdiagnose';
NEBENDIAGNOSE = 'DK'
NEW_FEATURES = ['previous_visits', 'ratio_los_age', 'ratio_numDK_age', 'ratio_los_numDK', 'ratio_numCHOP_age',
'ratio_los_numOE', 'ratio_numOE_age', 'mult_los_numCHOP', 'mult_equalOE_numDK',
'ratio_drg_los_alos'];
import sys
import shutil
import tensorflow as tf
from tensorflow.python.summary import summary
import helpers.helpers as helpers
diag_group_names = helpers.getDKverylightGrouping()
class AutoEncoderEstimator:
def __init__(self, feature_columns, flags):
self.feature_columns = feature_columns
self.flags = flags
self.estimator = None
return
def _add_hidden_layer_summary(self, value, tag):
summary.scalar('%s/fraction_of_zero_values' % tag,
tf.nn.zero_fraction(value))
summary.histogram('%s/activation' % tag, value)
def _dense_batch_relu(self, input, num_nodes, phase, layer_name, batchnorm,
dropout):
if batchnorm:
out = tf.layers.dense(input,
num_nodes,
activation=tf.nn.relu,
name=layer_name)
out = tf.layers.batch_normalization(out, training=phase)
dirNN = '/Users/towyku74/UniBas/sciCore/projects/PATREC/trained_models/dev/nz_20012011_reduction_FUSION_embedding_verylightgrouping_20_10_10_dropout_0.5_learningrate_0.05_batchnorm_True_batchsize_640/'
filename_weights_main_diag = dirNN + 'weights_embedding_main_diag.npy'
weights = np.load(filename_weights_main_diag)
num_diags = 2600
num_categories = 26
cnt = 0
labels = np.zeros(num_diags)
labels_maincat = []
for k in range(0, 26):
for l in range(0, 100):
labels[cnt] = k
labels_maincat.append(alphabet[k])
cnt += 1
labels_finegrained = helpers.getDKverylightGrouping()
filename_labels = dirNN + 'labels_cat.tsv'
file_labels = open(filename_labels, 'w')
file_labels.write('main_category' + '\t' + 'category' + '\n')
for k in range(0, len(labels_maincat)):
file_labels.write(labels_maincat[k] + '\t' + labels_finegrained[k] +
'\n')
file_labels.close()
colors = plt.cm.rainbow(np.linspace(0, 1, num_categories))
pca = PCA(n_components=2)
weights_2d_pca = pca.fit_transform(weights)
tsne = TSNE(n_components=2)
weights_2d_tsne = tsne.fit_transform(weights)