antgo 0.0.14

machine learning experiment platform

Target

Antgo is a machine learning experiment manage platform, which has been integrated deeply with MLTalker. Antgo provides some easy cli commands to help ML researchers to manage, analyze, and challenge all kinds of ML tasks.

Based on amounts of statistical evaluation methods, Antgo could give a fruitful evaluation report, which help researchers analyze and trade-off their model.

Antgo tutorial is at MLTalker Blog.

Installation

1.step install Antgo:

pip install antgo

or install Antgo from source:

1. git clone https://github.com/jianzfb/antgo.git
2. cd antgo
3. pip install -r requirements.txt
4. python setup.py build_ext install

2.step generate config file and modify:

antgo config

in current folder, you will get:

config.xml

content in config.xml is like this:

<?xml version='1.0' encoding='utf-8'?>
<antgo>
    <factory></factory>
    <server_ip>www.mltalker.com</server_ip>
    <server_port>8999</server_port>
    <server_user_token></server_user_token>
</antgo>

In <factory> tag, fill your local folder. In this folder, you should build two subfolders

• dataset

  all data should put here

• task

  all task file should put here

In <server_user_token> tag, fill your user api-token. How to get your api-token, see next section.

3.step update config file:

antgo config --config=./config.xml

Register

Register in MLTalker.

All user experiment records would be managed by MLTalker in user’s personal page.

Quick Example

1.step Build and Apply Task

Build and Apply Task in

2.step Create Your Project

Create Your project

antgo startproject –name=MNIST –author=xxx –token=Task API-TOKEN

after that, you will get in current folder

3.step write your train and predict code

in MNIST_main.py, you should finish training_callback and infer_callback functions.

training_callback function:

def training_callback(data_source, dump_dir):
    # warning: data_source include data and label
    try:
        # data_source.iterator_value() get generator
        for index, (data, label) in enumerate(data_source.iterator_value()):
            # data, label is data and its label
            pass
    except:
        pass

    # build logger to record important data
    mc = mlogger.Container()
    mc.loss = mlogger.metric.Simple('model loss')

    epochs = 100
    for epoch in range(epochs):
        for _ in range(500):
            # train model
            ...
            # loss value
            loss_val = ...
            mc.loss.update(loss_val)

        # save best model
        ...

infer_callback function:

def infer_callback(data_source, dump_dir):
    # warning: dont include label data
    # get dataset size
    data_size = data_source.size
    # parse data
    try:
        # data_source.iterator_value() get generator
        for index, data in enumerate(data_source.iterator_value()):
            pass
    except:
        pass

    # load from model from ctx.from_experiment
    # ctx.from_experiment is experiment_uuid (shell script)

    # run predict
    ...

    # record every sample predict result
    for index in range(data_size):
        ctx.recorder.record({
          'RESULT': (int)(score[index])
        })

you can go MLTalker Blog, to see more cases.

4.step Run Train Task

antgo train exp

5.step Run Challenge Task

antgo challenge exp experiment_uuid