def run_gene_model():
train_cm_X = train_cm.iloc[:, 0:-ADDITIONAL_COLUMN].values
train_cm_Y = train_cm.loc[:, LABEL_COLUMN].values
train_cm_Y_one_hot = to_categorical(label_encoder.transform(train_cm_Y),
num_classes=len(class_list))
test_cm_X = test_cm.iloc[:, 0:-ADDITIONAL_COLUMN].values
test_cm_Y = test_cm.loc[:, LABEL_COLUMN].values
test_cm_Y_one_hot = to_categorical(label_encoder.transform(test_cm_Y),
num_classes=len(class_list))
parallel_model_gene = st_nn(train_cm_X.shape[1],
train_cm_Y_one_hot.shape[1])
parallel_model_gene.fit(train_cm_X,
train_cm_Y_one_hot,
batch_size=batch_size,
epochs=epochs)
y_pred = parallel_model_gene.predict(test_cm_X, verbose=1)
# get true label (categorical)
y_true = test_cm_Y
# convert categorical string label to numerical label
y_true = label_encoder.transform(y_true)
# plot ROC and confusion matrix
out_path = os.path.join(DATASET_PATH, "results_gene_model")
mkdirs(out_path)
model_eval(y_pred,
y_true,
class_list=class_list,
prefix=os.path.join(out_path, "single_model_gene_count"))
def prediction():
filename = request.args.get('filename')
pred, probs = model_eval(
os.path.join(app.config['UPLOAD_FOLDER'], filename))
probs = sorted(list(probs.items()), key=lambda x: x[1], reverse=True)
return render_template('display.html',
filename=filename,
prediction=pred,
probabilities=probs)
def get_eval_op():
dataset = DataGenerator(
sample_range=config.eval_samples,
shuffle_samples=True,
max_patches_per_sample=config.max_patches_per_sample
).make_dataset(
split_lhs_rhs=False
)
X, Y = dataset.make_one_shot_iterator().get_next()
ops = model_eval(X, Y, params={'n_candidates': tf.constant(config.n_candidates)})
return ops
def run_combine_model():
# create input batch for training model. eg. ([image, cm], label) * batch_size
train_gen = DataGenerator(df=df,
cm_df=train_cm,
le=label_encoder,
cm_len=cm_shape,
batch_size=batch_size)
# create test data generator. eg. [image, cm] * 1
test_gen = DataGenerator(df=df,
cm_df=test_cm,
le=label_encoder,
shuffle=False,
is_train=False,
batch_size=1,
cm_len=cm_shape)
# compile model
parallel_model_combine, model_ = st_comb_nn((SIZE[0], SIZE[1], N_CHANNEL),
(cm_shape, ), len(class_list))
# training model
parallel_model_combine.fit_generator(generator=train_gen,
steps_per_epoch=len(train_gen),
epochs=epochs)
# predict on test dataset to get predict values with shape (n_classes,)
y_pred = parallel_model_combine.predict_generator(generator=test_gen,
verbose=1)
# get true label (categorical)
y_true = test_cm.loc[:, LABEL_COLUMN].values
# convert categorical string label to numerical label
y_true = label_encoder.transform(y_true)
# plot ROC and confusion matrix
out_path = os.path.join(DATASET_PATH, "results_combine_model")
mkdirs(out_path)
model_eval(y_pred,
y_true,
class_list=class_list,
prefix=os.path.join(out_path, "combine_model"))
def eval():
global net
data = request.get_json()
image = Image.open(io.BytesIO(base64.b64decode(data["data"]))).convert("RGB")
return {"data": model_eval(net, image)}