load_multiple_keras_models_django.py

load_multiple_keras_models_django.py

###################################################
# settings.py/the file where to load the models
###################################################
def load_model_from_path(path):
    graph = tf.get_default_graph()
    model = load_model(path)
    return graph, model

def load_all_models():
    global gModelObjs # each object is a tuple of graph, model
    gModelObjs = dict()

    gModelObjs = {
        'model_1': load_model_from_path('path_for_model_1'),
        'model_2': load_model_from_path('path_for_model_2'),
        'model_3': load_model_from_path('path_for_model_3'),
        'model_4': load_model_from_path('path_for_model_4'),
    }

#########
# view.py
#########

@api_view(['POST'])
def score_segment(request):
    # We no longer need to load the model here. It's already preloaded.
    # graph, model = _load_model_from_path('path_to_keras_model')
    graph, model = gModelObjs('model_1') # if model_1 is used
    df_segment = pd.DataFrame(request.data)
    try:
        predictions = _score_segment_by_model(df_segment, graph, model)
    except Exception as error:
        return Response({'message':str(error)}, status=status.HTTP_400_BAD_REQUEST)
    return Response({
        'predictions': predictions
    })