skt 0.2.101

SKT package

SKT Package

This is highly site dependent package. Resources are abstracted into package structure.

Usage

Hive metastore

from skt.ye import get_hms

c = get_hms()
c.get_partition_names("db", "table")
c.close()

Hash and unhash

from skt.lake import hash_s
from skt.lake import unhash_s

unhashed_list = ['0000000000']
hashed_list = hash_s(unhashed_list)
unhash_s(hashed_list)

Execute hive query without fetch result

from skt.ye import hive_execute
hive_execute(ddl_or_ctas_query)

Fetch resultset from hive query

from skt.ye import hive_get_result
result_set = hive_get_result(select_query)

Get pandas dataframe from hive qeruy resultset

from skt.ye import hive_to_pandas
pandas_df = hive_to_pandas(hive_query)

Get pandas dataframe from parquet file in hdfs

from skt.ye import parquet_to_pandas
pandas_df = parquet_to_pandas(hdfs_path)

Save pandas dataframe as parquet in hdfs

from skt.ye import get_spark
from skt.ye import pandas_to_parquet
spark = get_spark()
pandas_to_parquet(pandas_df, hdfs_path, spark) # we need spark for this operation
spark.stop()

Work with spark

from skt.ye import get_spark
spark = get_spark()
# do with spark session
spark.stop()

Work with spark-bigquery-connector

# SELECT
from skt.gcp import bq_table_to_pandas 
pandas_df = bq_table_to_pandas("dataset", "table_name", ["col_1", "col_2"], "2020-01-01", "svc_mgmt_num is not null")
# INSERT 
from skt.gcp import pandas_to_bq_table
pandas_to_bq_table(pandas_df, "dataset", "table_name", "2020-03-01")

Send slack message

from skt.ye import slack_send
text = 'Hello'
username = 'airflow'
channel = '#leavemealone'
slack_send(text=text, username=username, channel=channel)
# Send dataframe as text
df = pd.DataFrame(data={'col1': [1, 2], 'col2': [3, 4]})
slack_send(text=df, username=username, channel=channel, dataframe=True)

Get bigquery client

from skt.gcp import get_bigquery_client
bq = get_bigquery_client()
bq.query(query)

IPython BigQuery Magic

from skt.gcp import import_bigquery_ipython_magic

import_bigquery_ipython_magic()

query_params = {
    "p_1": "v_1",
    "dataset": "mnoai",
}

%% bq --params $query_params

SELECT c_1 
FROM {dataset}.user_logs
WHERE c_1 = @p_1

Access MLS

from skt.mls import set_model_name
from skt.mls import get_recent_model_path
from skt.ye import get_pkl_from_hdfs

set_model_name(COMM_DB, params)
path = get_recent_model_path(COMM_DB, model_key)
model = get_pkl_from_hdfs(f'{path})

MLS Model Registry (Upload model_binary(model.tar.gz) / model_meta(model.json) to AWS S3 from YE)

from skt.mls import save_model

# model object generated by LightGBM or XGBoost
model

# model name
model_name = 'sample_model'
# model version
model_version = 'v1'
# AWS ENV in 'stg / prd / dev' (default is 'stg')
aws_env = 'stg'
# List of features used in ML Model in string type (only for XGBoost model_type)
feature_list = ['feature_1', 'feature_2', 'feature_3']
# Force to overwrite model files on S3 if exists (default is False)
force = False 

save_model(model, model_name, model_version, aws_env, force)

MLS meta_table & meta_table_item related methods

from skt.mls import get_meta_table
from skt.mls import create_meta_table_item
from skt.mls import update_meta_table_item
from skt.mls import get_meta_table_item
from skt.mls import meta_table_to_pandas
from skt.mls import pandas_to_meta_table

# Get a meta_table info
get_meta_table(meta_table_name, aws_env, edd)
# Create a meta_item
create_meta_table_item(meta_table_name, item_name, item_dict, aws_env, edd)
# Update a meta_item
update_meta_table_item(meta_table_name, item_name, item_dict, aws_env, edd)
# Get a meta_item
get_meta_table_item(meta_table_name, item_name, aws_env, edd)
# Get a meta_table as pandas dataframe
meta_table_to_pandas(meta_table_name, aws_env, edd)
# Update pandas dataframe to meta_table
pandas_to_meta_table(method, meta_table_name, dataframe, key, values, aws_env, edd)


# For the detal, use ?{method} to get detailed info (ex. ?get_meta_table)
# For the user of EDD, must set edd=True

MLS model_meta related methods
(*Need to set user for the ml_model)

from skt.mls import get_ml_model
from skt.mls import create_meta_table_item
from skt.mls import update_meta_table_item

# Get a ml_model
get_ml_model(user, model_name, model_version, aws_env, edd)
# Get a model_meta of ml_model
get_ml_model_meta(user, model_name, model_version, aws_env, edd)
# Update or Create meta_item(s)
update_ml_model_meta(user, model_name, model_version, model_meta_dict, aws_env, edd)

# For the detal, use ?{method} to get detailed info (ex. ?get_ml_model)
# For the user of EDD, must set edd=True

Use NES CLI

nes input_notebook_url -p k1 v1 -p k2 v2 -p k3 v3

Use github util

from skt.ye import get_github_util
g = get_github_util
# query graphql
res = g.query_gql(graph_ql)
# get file in github repository
byte_object = g.download_from_git(github_url_path)

Installation

$ pip install skt --upgrade

If you would like to install submodules for AIR

$ pip install skt[air] --upgrade

Develop

Create issue first and follow the GitHub flow https://help.github.com/en/github/collaborating-with-issues-and-pull-requests/github-flow

AIPS EDA tools

OVERVIEW

• Modeling EDA 시 활용할 수 있는 기능의 공통 module
• Modules
  • 1. EDA (Nuemric / Categorical variable)

1) EDA

1. Numeric variable EDA

• def numeric_eda_plot

    Numeric feature에 대한 EDA Plot function

    Args. :
        - df           :   Pandas DataFrame 형태의 EDA대상 데이터
        - feature_list :   EDA 대상 feature list (df의 columns)
        - label_col    :   Label(or Hue) column
        - cols         :   Multi-plot 시 grid column 개수 (row 개수는 feature_list에 따라 자동으로 결정 됨)
        - n_samples    :   Label 별 sampling 할 개수 (default = -1(전수 데이터로 EDA할 경우))
        - plot_type    :   density or box (default = 'density')
        - stat_yn      :   기초 통계량 출력여부 (mean / min / max / 1q / 3q) (default : False)
        - figsize      :   (default : (7,4))

    Returns : 
        matplotlib.pyplot object

    Example : 
        fig = numeric_eda_plot(df, ['age'], 'answer', cols = 1, n_samples = 10000, plot_type='density', stat_yn=True, figsize = (7,4))
        fig

        if want to Save the EDA images,
        fig.savefig('filename')

2. Categorical variable EDA

• def categorical_eda_plot

    Categorical feature에 대한 EDA Plot function

    Args. :
        - df           :   Pandas DataFrame 형태의 EDA대상 데이터
        - feature_list :   EDA 대상 feature list (df의 columns)
        - label_col    :   Label(or Hue) column
        - cols         :   Multi-plot 시 grid column 개수 (row 개수는 feature_list에 따라 자동으로 결정 됨)
        - n_samples    :   Label 별 sampling 할 개수 (default = -1(전수 데이터로 EDA할 경우))
        - figsize      :   (default : (7,4))

    Returns : 
        matplotlib.pyplot object


    Example : 
        Example : 
        fig = categorical_eda_plot(df, ['sex_cd'], 'answer', cols = 1, n_samples = 10000, figsize = (7,4))
        fig

        if want to Save the EDA images,
        fig.savefig('filename')