Endpoint API

Default function

• load_model

• predict

• predict_file

• on_train_completed

load_model

load model and checkpoint here

Example:

class Model:
    ...

def load_model():
    model = Model()
    model.load_state_dict(torch.load(modelLoader.path))
    model.to(device)
    model.eval()
    modelLoader.update_model(model)

predict(data)

predict with http post body

Arguments
data	Dict (http post body)

Example:

def predict(data):
    model = modelLoader.get_model()
    files = magic.download(data['images'])
    predict = []
    file_name = []
    for file in files:
        img, label = load_image(file)
        result = model(img)
        _, pred = torch.max(result, 1)
        predict.append(int(pred))
        file_name.append(label)
    result = {
        'data': {
            'predict': predict,
            'file_name': file_name
        }
    }
    return result

predict_file(files)

predict with http post file

Arguments
files	File (http post file)

Example:

def predict_file(files):
    model = modelLoader.get_model()
    files = magic.save(files)
    predict = []
    file_name = []
    for file in files:
        img, label = load_image(file)
        result = model(img)
        _, pred = torch.max(result, 1)
        predict.append(int(pred))
        file_name.append(label)
    result = {
        'data': {
            'predict': predict,
            'file_name': file_name
        }
    }
    return result

on_train_completed(metric, config)

Will call after the experiment is completed

Arguments
metric	Float (this experiment metric)
config	Dict (this experiment best conf9ig)

Example:

def on_train_completed(metric, config):
    if metric > pipelineHepler.last_metric:
        modelLoader.save_model()
        load_model()
        pipelineHepler.update_metric(metric)