Luk.ai Python管理库

这是一个用于将机器学习模型上传到Luk.ai的库。

上传模型

您首先需要 创建API令牌。

import lukai

# ... your model definition code

sess = tf.Session()
sess.run(tf.initialize_all_variables())

# Sets the Luk.ai API token.
lukai.set_api_token('<your token>')

# Uploads the model to Luk.ai and creates a training job.
lukai.upload(
    session=sess,
    domain='<your domain>',
    model_type='<your model type>',
    name='Hello World',
    description='This is the first model I've uploaded!',
    hyper_params=lukai.HyperParams(
        num_clients = 10,
        batch_size = 10,
        num_rounds = 100,
        learning_rate = learning_rate,
        num_local_rounds = 10,
    ),
    metrics={
      accuracy: lukai.REDUCE_MEAN,
    },
    event_targets={
      lukai.EVENT_TRAIN: (keep_prob.assign(0.5),),
      lukai.EVENT_INFER: (keep_prob.assign(1.0),),
      lukai.EVENT_EVAL: (keep_prob.assign(1.0),),
    },
)

查看完整的 mnist示例。

导出模型

如果您愿意，也可以直接输出 model.tar.gz 文件。

from lukai import saver

# ... your model definition code

sess = tf.Session()
sess.run(tf.initialize_all_variables())

print('Node names: x = {}, y_ = {}, train_step = {}, w = {}, b = {}, y = {}'.format(
  x.name, y_.name, train_step.name, w.name, b.name, y.name,
))

saver.save(sess)

查看完整的 leastsquares示例