def evaluate(start_date, end_date, all_dtoc=False):
df = dtoc.sample(40000, start_date, end_date)
if all_dtoc:
df = df[df['is_dtoc'] == 0]
sample_x = vectorization(df.iloc[:, 4:16], Word2Vec.load("diag2vec.model"))
model = load_model()
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=[f1, 'acc'])
return model.evaluate(sample_x, df['is_dtoc'])
def train():
## load data
df = dtoc.sample(40000)
train_df, val_df = train_test_split(df, test_size=0.08, random_state=2018)
train_X = train_df[[
'diag1', 'diag2', 'diag3', 'diag4', 'diag5', 'diag6', 'diag7', 'diag8',
'diag9', 'diag10', 'diag11', 'diag12', 'age'
]]
# train_X = train_df['age'].values
train_y = train_df['is_dtoc'].values
val_X = val_df[[
'diag1', 'diag2', 'diag3', 'diag4', 'diag5', 'diag6', 'diag7', 'diag8',
'diag9', 'diag10', 'diag11', 'diag12', 'age'
]]
# val_X = train_df['age'].values
val_y = val_df['is_dtoc'].values
# embed_size = 150 # how big is each variable vector
# max_features = 10000 # how many unique codes to use (ICD codes + all ages)
# max_len = 13 # max number of variables in one records
# embedding_matrix = get_embeddings_matrix(df)
# load word2vec model
tbCallBack = TensorBoard(log_dir='./Graph',
histogram_freq=0,
write_graph=True,
write_images=True)
EMBEDDING_MODEL_FILE = 'diag2vec.model'
wv_model = Word2Vec.load(EMBEDDING_MODEL_FILE)
#model = model_LR()
model = model_LSTM()
# model = model_CNN()
model.fit(vectorization(train_X, wv_model),
train_y,
batch_size=512,
epochs=20,
validation_data=(vectorization(val_X, wv_model), val_y),
callbacks=[tbCallBack])
pred_val_y = model.predict(vectorization(val_X, wv_model),
batch_size=1024,
verbose=1)
for thresh in np.arange(0.1, 0.501, 0.01):
thresh = np.round(thresh, 2)
print("F1 score at threshold {0} is {1}".format(
thresh, metrics.f1_score(val_y,
(pred_val_y > thresh).astype(int))))
def main():
df = dtoc.sample(40000, '2010-01-01', '2018-01-01')
train_df, val_df = train_test_split(df, test_size=0.1)
wv_model = Word2Vec.load("diag2vec.model")
validation_data = (vectorization(val_df.iloc[:, 4:16],
wv_model), np.array(val_df['is_dtoc']))
model = train(train_df, wv_model, validation_data)
##save weights
model.save_weights('lstm_weights_before2018.h5')
##save structure
with open('lstm_architecture_before2018.json', 'w') as f:
f.write(model.to_json())
def predict_evaluate():
model = load_model()
df = dtoc.sample(40000)
val_y = df['is_dtoc']
sample_x = vectorization(df.iloc[:, 4:16], Word2Vec.load("diag2vec.model"))
model = load_model()
# model.compile(loss='binary_crossentropy',optimizer='adam',metrics=[f1, 'acc'])
pred_val_y = model.predict(sample_x)
thresholds = []
for thresh in np.arange(0.1, 0.501, 0.01):
thresh = np.round(thresh, 2)
res = metrics.f1_score(val_y, (pred_val_y > thresh).astype(int))
thresholds.append([thresh, res])
print("F1 score at threshold {0} is {1}".format(thresh, res))