PdpCLI

Quick Links

• Introduction
• Installation
• Tutorial
  • Basic Usage
  • Data Reader / Writer
  • Plugins

Introduction

PdpCLI is a pandas DataFrame processing CLI tool which enables you to build a pandas pipeline powered by pdpipe from a configuration file. You can also extend pipeline stages and data readers / writers by using your own python scripts.

Features

• Process pandas DataFrame from CLI without wrting Python scripts
• Support multiple configuration file formats: YAML, JSON, Jsonnet
• Read / write data files in the following formats: CSV, TSV, JSON, JSONL, pickled DataFrame
• Import / export data with multiple protocols: S3 / Databse (MySQL, Postgres, SQLite, ...) / HTTP(S)
• Extensible pipeline and data readers / writers

Installation

Installing the library is simple using pip.

$ pip install "pdpcli[all]"

Tutorial

Basic Usage

1. Write a pipeline config file config.yml like below. The type fields under pipeline correspond to the snake-cased class names of the PdpipelineStages. Other fields such as stage and columns are the parameters of the __init__ methods of the corresponging classes. Internally, this configuration file is converted to Python objects by colt.

pipeline:
  type: pipeline
  stages:
    drop_columns:
      type: col_drop
      columns:
        - name
        - job

    encode:
      type: one_hot_encode
      columns: sex

    tokenize:
      type: tokenize_text
      columns: content

    vectorize:
      type: tfidf_vectorize_token_lists
      column: content
      max_features: 10

2. Build a pipeline by training on train.csv. The following command generages a pickled pipeline file pipeline.pkl after training. If you specify a URL of file path, it will be automatically downloaded and cached.

$ pdp build config.yml pipeline.pkl --input-file https://github.com/altescy/pdpcli/raw/main/tests/fixture/data/train.csv

3. Apply the fitted pipeline to test.csv and get output of a processed file processed_test.jsonl by the following command. PdpCLI automatically detects the output file format based on the file name. In this example, the processed DataFrame will be exported as the JSON-Lines format.

$ pdp apply pipeline.pkl https://github.com/altescy/pdpcli/raw/main/tests/fixture/data/test.csv --output-file processed_test.jsonl

4. You can also directly run the pipeline from a config file without fitting pipeline.

$ pdp apply config.yml test.csv --output-file processed_test.jsonl

5. It is possible to override or add parameters by adding command line arguments:

pdp apply config.yml test.csv pipeline.stages.drop_columns.column=name

Data Reader / Writer

PdpCLI automatically detects a suitable data reader / writer based on a given file name. If you need to use the other data reader / writer, add a reader or writer config to config.yml. The following config is an exmaple to use SQL data reader. SQL reader fetches records from the specified database and converts them into a pandas DataFrame.

reader:
    type: sql
    dsn: postgres://${env:POSTGRES_USER}:${env:POSTGRES_PASSWORD}@your.posgres.server/your_database

Config files are interpreted by OmegaConf, so ${env:...} is interpolated by environment variables.

Prepare your SQL file query.sql to fetch data from the database:

select * from your_table limit 1000

You can execute the pipeline with SQL data reader via:

$ POSTGRES_USER=user POSTGRES_PASSWORD=password pdp apply config.yml query.sql

Plugins

By using plugins, you can extend PdpCLI. This plugin feature enables you to use your own pipeline stages, data readers / writers and commands.

Add a new stage

1. Write your plugin script mypdp.py like below. Stage.register("<stage-name>") registers your pipeline stages, and you can specify these stages by writing type: <stage-name> in your config file.

import pdpcli

@pdpcli.Stage.register("print")
class PrintStage(pdpcli.Stage):
    def _prec(self, df):
        return True

    def _transform(self, df, verbose):
        print(df.to_string(index=False))
        return df

2. Update config.yml to use your plugin.

pipeline:
    type: pipeline
    stages:
        drop_columns:
        ...

        print:
            type: print

        encode:
        ...

2. Execute command with --module mypdp and you can see the processed DataFrame after running drop_columns.

$ pdp apply config.yml test.csv --module mypdp

Add a new command

You can also add new commands not only stages.

1. Add the following script to mypdp.py. This greet command prints out a greeting message with your name.

@pdpcli.Subcommand.register(
    name="greet",
    description="say hello",
    help="say hello",
)
class GreetCommand(pdpcli.Subcommand):
    requires_plugins = False

    def set_arguments(self):
        self.parser.add_argument("--name", default="world")

    def run(self, args):
        print(f"Hello, {args.name}!")

2. To register this command, you need to create the .pdpcli_plugins file in which module names are listed for each line. Due to module importing order, the --module option is unavailable for command registration.

$ echo "mypdp" > .pdpcli_plugins

3. Run the following command and get a message like below. By using the .pdpcli_plugins file, it is is not needed to add the --module option to a command line for each execution.

$ pdp greet --name altescy
Hello, altescy!