Example

Upload, Setup, Train, Predict

Import Packages:

pip install decanter-ai-core-sdk

import os
import asyncio

from decanter import core
from decanter.core.core_api import TrainInput, PredictInput, SetupInput
from decanter.core.extra.utils import check_response, gen_id
from decanter.core.enums.algorithms import Algo
from decanter.core.enums.evaluators import Evaluator

Create Context will set the connection to decanter core server, and create an event loop. Since Jupyter already have an event loop, SDK will just use the current event loop. In Jupyter, it will initially exist a running event loop.:

loop = asyncio.get_running_loop()
loop.is_running()

CoreClient handles the actions of calling api and getting the results,: When initializing, need to set the usr, pwd, host to create Context.:

# enable default logger
core.enable_default_logger()
# set the username, password, host
client = core.CoreClient(
        username=???, password=???, host='http://host:port')

Open train & test file:

train_file_path = '/data/train.csv'
test_file_path = '/data/test.csv'
train_file = open(train_file_path , 'rb')
test_file = open(test_file_path , 'rb')

Upload data to CoreX:

train_data = client.upload(file=train_file, name="train_data")
test_data = client.upload(file=test_file, name="test_data")

Setup data to CoreX:

    setup_input = SetupInput(
    data = train_data,
    data_source=train_data.accessor,
    data_columns=[
        {
            'id': 'Pclass',
            'data_type': 'categorical'
        }])
train_data = client.setup(setup_input=setup_input, name='setup_data')

Set train parameters train model:

train_input = TrainInput(data=train_data, target='Survived', algos=[Algo.XGBoost], max_model=2, tolerance=0.9)
exp = client.train(train_input=train_input, select_model_by=Evaluator.mean_per_class_error, name='myexp')

Set predict parameters and predict result:

predict_input = PredictInput(data=test_data, experiment=exp)
pred_res = client.predict(predict_input=predict_input, name='mypred')

Show the predict result:

pred_res.show_df()

How to Save Model

Getting Mojo model zip file from decanter.core server and download to local.

from decanter.core.core_api import Model
model = Model()

"""
save the model as zip file
    model_id (str): ObjectId in 24 hex digits.
    model_path (str): Path to store zip mojo file.
"""
model_id = {model_id}
model_path = './tmp/model.zip'
model.download_by_id(model_id, model_path)