Skip to main content

Optimus is the missing framework for cleaning and pre-processing data in a distributed fashion with pyspark.

Project description

Logo Optimus

PyPI version Build Status Documentation Status built_by iron Updates GitHub release Codacy Badge Coverage Status Mentioned in Awesome Data ScienceDiscord

Downloads Downloads Downloads

NOTE: We are working hard in version 3.0 in which you could use pandas, cudf, Dask, Dask/cudf as backend using a unified API.

To launch a live notebook server to test optimus using binder or Colab, click on one of the following badges:

Binder Colab

Optimus is the missing framework to profile, clean, process and do ML in a distributed fashion using Apache Spark(PySpark).

Installation (pip):

In your terminal just type pip install optimuspyspark


  • Apache Spark>= 2.4.0
  • Python>=3.6


You can go to the 10 minutes to Optimus notebook where you can find the basic to start working.

Also you can go to the examples folder to found specific notebooks about data cleaning, data munging, profiling, data enrichment and how to create ML and DL models.

Besides check the Cheat Sheet




Feedback is what drive Optimus future, so please take a couple of minutes to help shape the Optimus' Roadmap:

Also if you want to a suggestion or feature request use

Start Optimus

from optimus import Optimus
op= Optimus(verbose=True)

You also can use an already created Spark session:

from pyspark.sql import SparkSession
from optimus import Optimus

spark = SparkSession.builder.appName('optimus').getOrCreate()
op= Optimus(spark)

Loading data

Now Optimus can load data in csv, json, parquet, avro, excel from a local file or URL.

df = op.load.csv("../examples/data/foo.csv")

# Use a local file
df = op.load.json("../examples/data/foo.json")

# Use a url
df = op.load.json("")

# parquet
df = op.load.parquet("../examples/data/foo.parquet")

# avro
# df = op.load.avro("../examples/data/foo.avro").table(5)

# excel 
df = op.load.excel("../examples/data/titanic3.xls")

Also you can load data from oracle, redshift, mysql and postgres. See Database connection

Saving Data


# json"data/foo.json")

# parquet"data/foo.parquet")

# avro"examples/data/foo.avro")

Also you can save data to oracle, redshift, mysql and postgres. See Database connection

Handling Spark jars, packages and repositories

With optimus is easy to loading jars, packages and repos. You can init optimus/spark like

op= Optimus(repositories = "myrepo", packages="org.apache.spark:spark-avro_2.12:2.4.3", jars="my.jar", driver_class_path="this_is_a_jar_class_path.jar", verbose= True)

Create dataframes

Also you can create a dataframe from scratch

from pyspark.sql.types import *
from datetime import date, datetime

df = op.create.df(
        ("names", "str", True), 
        ("height(ft)","int", True), 
        ("function", "str", True), 
        ("rank", "int", True), 
        ("japanese name", ArrayType(StringType()), True),
        ("last position seen", "str", True),
        ("date arrival", "str", True),
        ("last date seen", "str", True),
        ("attributes", ArrayType(FloatType()), True),
        ("Cybertronian", "bool", True), 
        ("NullType", "null", True),
        ("Optim'us", 28, "Leader", 10, 5000000, 4.3, ["Inochi", "Convoy"], "19.442735,-99.201111", "1980/04/10",
         "2016/09/10", [8.5344, 4300.0], date(2016, 9, 10), datetime(2014, 6, 24), True,
        ("bumbl#ebéé  ", 17, "Espionage", 7, 5000000, 2.0, ["Bumble", "Goldback"], "10.642707,-71.612534", "1980/04/10",
         "2015/08/10", [5.334, 2000.0], date(2015, 8, 10), datetime(2014, 6, 24), True,
        ("ironhide&", 26, "Security", 7, 5000000, 4.0, ["Roadbuster"], "37.789563,-122.400356", "1980/04/10",
         "2014/07/10", [7.9248, 4000.0], date(2014, 6, 24), datetime(2014, 6, 24), True,
        ("Jazz", 13, "First Lieutenant", 8, 5000000, 1.80, ["Meister"], "33.670666,-117.841553", "1980/04/10",
         "2013/06/10", [3.9624, 1800.0], date(2013, 6, 24), datetime(2014, 6, 24), True, None),
        ("Megatron", None, "None", 10, 5000000, 5.70, ["Megatron"], None, "1980/04/10", "2012/05/10", [None, 5700.0],
         date(2012, 5, 10), datetime(2014, 6, 24), True, None),
        ("Metroplex_)^$", 300, "Battle Station", 8, 5000000, None, ["Metroflex"], None, "1980/04/10", "2011/04/10",
         [91.44, None], date(2011, 4, 10), datetime(2014, 6, 24), True, None),

    ], infer_schema = True).h_repartition(1)

With .table() you have a beautifull way to show your data. You have extra information like column number, column data type and marked white spaces


Also you can create a dataframe from a panda dataframe

import pandas as pd
pdf = pd.DataFrame({'A': {0: 'a', 1: 'b', 2: 'c',3:'d'},
                    'B': {0: 1, 1: 3, 2: 5,3:7},
                       'C': {0: 2, 1: 4, 2: 6,3:None},
                       'D': {0:'1980/04/10',1:'1980/04/10',2:'1980/04/10',3:'1980/04/10'},

s_pdf = op.create.df(pdf=pdf)

Cleaning and Processing

Optimus V2 was created to make data cleaning a breeze. The API was designed to be super easy to newcomers and very familiar for people that comes from Pandas. Optimus expands the Spark DataFrame functionality adding .rows and .cols attributes.

For example you can load data from a url, transform and apply some predefined cleaning functions:

# This is a custom function
def func(value, arg):
    return "this was a number"

new_df = df\
    .withColumn('new_age', df.age)\
    .cols.date_transform("date arrival", "yyyy/MM/dd", "dd-MM-YYYY")\
    .cols.years_between("date arrival", "dd-MM-YYYY", output_cols = "from arrival")\
    .cols.unnest("japanese name", output_cols="other names")\
    .cols.unnest("last position seen",separator=",", output_cols="pos")\
    .cols.drop(["last position seen", "japanese name","date arrival", "cybertronian", "nulltype"])

You transform this


Into this


Note that you can use Optimus functions and Spark functions(.WithColumn()) and all the df function availables in a Spark Dataframe at the same time. To know about all the Optimus functionality please go to this notebooks

Handling column output

With Optimus you can handle how the output column from a transformation in going to be handled.

from pyspark.sql import functions as F

def func(col_name, attr):
    return F.upper(F.col(col_name))

If a string is passed to input_cols and output_cols is not defined the result from the operation is going to be saved in the same input column

output_df = df.cols.apply(input_cols="names", output_cols=None,func=func)

If a string is passed to input_cols and a string is passed to output_cols the output is going to be saved in the output column

output_df = df.cols.apply(input_cols="names", output_cols="names_up",func=func)

If a list is passed to input_cols and a string is passed to out_cols Optimus will concatenate the list with every element in the list to create a new column name with the output

output_df = df.cols.apply(input_cols=["names","function"], output_cols="_up",func=func)

If a list is passed to input_cols and a list is passed in out_cols Optimus will output every input column in the respective output column

output_df = df.cols.apply(input_cols=["names","function"], output_cols=["names_up","function_up"],func=func)

Custom functions

Spark has multiple ways to transform your data like rdd, Column Expression, udf and pandas udf. In Optimus we created the apply() and apply_expr which handles all the implementation complexity.

Here you apply a function to the "billingid" column. Sum 1 and 2 to the current column value. All powered by Pandas UDF

def func(value, args):
    return value + args[0] + args[1]

df.cols.apply("height(ft)",func,"int", [1,2]).table()

If you want to apply a Column Expression use apply_expr() like this. In this case we pass an argument 10 to divide the actual column value

from pyspark.sql import functions as F

def func(col_name, args):
    return F.col(col_name)/20

df.cols.apply("height(ft)", func=func, args=20).table()

You can change the table output back to ascii if you wish


To return to HTML just:


Data profiling

Optimus comes with a powerful and unique data profiler. Besides basic and advance stats like min, max, kurtosis, mad etc, it also let you know what type of data has every column. For example if a string column have string, integer, float, bool, date Optimus can give you an unique overview about your data. Just run df.profile("*") to profile all the columns. For more info about the profiler please go to this notebook.

Let's load a "big" dataset

df = op.load.csv("").h_repartition()

Numeric, "mass (g)", infer=False), "name", infer=False)

Processing Dates

For dates data types Optimus can give you extra information, "year", infer=True)

Profiler Speed

With relative_error and approx_count params you can control how some operations are caculated so you can speedup the profiling in case is needed.

relative_error: Relative Error for quantile discretizer calculation. 1 is Faster, 0 Slower

approx_count: Use approx_count_distinct or countDistinct. approx_count_distinct is faster, "mass (g)", infer=False, relative_error =1, approx_count=True)


Besides histograms and frequency plots you also have scatter plots and box plots. All powered by Apache by pyspark

df = op.load.excel("../examples/data/titanic3.xls")
df = df.rows.drop_na(["age","fare"])

You can output to the notebook or as an image

# Output and image


df.plot.scatter(["fare", "age"], buckets=30)"age")


Using other plotting libraries

Optimus has a tiny API so you can use any plotting library. For example, you can use df.cols.scatter(), df.cols.frequency(), df.cols.boxplot() or df.cols.hist() to output a JSON that you can process to adapt the data to any plotting library.


Get the ouliers using tukey


Remove the outliers using tukey



You can also use z_score, modified_z_score or mad

df.outliers.z_score("age", threshold=2).drop()
df.outliers.modified_z_score("age", threshold = 2).drop()
df.outliers.mad("age", threshold = 2).drop()

Database connection

Optimus have handy tools to connect to databases and extract informacion. Optimus can handle redshift, postgres, oracle and mysql

from optimus import Optimus
op= Optimus(verbose=True)
# This import is only to hide the credentials
from credentials import *

# For others databases use in db_type accepts 'oracle','mysql','redshift','postgres'

db =  op.connect(
    database= DATABASE,
    user= USER,
    password = PASSWORD,

# Show all tables names
# # Show a summary of every table"*",20)
# # Get a table as dataframe
df_ = db.table_to_df("places_interest").table()
# # Create new table in the database
db.df_to_table(df, "new_table")

Data enrichment

You can connect to any external API to enrich your data using Optimus. Optimus uses MongoDB to download the data and then merge it with the Spark Dataframe. You need to install MongoDB

Let's load a tiny dataset we can enrich

df = op.load.json("")
import requests

def func_request(params):
    # You can use here whatever header or auth info you need to send. 
    # For more information see the requests library

    url= "" + str(params["id"])
    return requests.get(url)

def func_response(response):
    # Here you can parse de response
    return response["title"]

e = op.enrich(host="localhost", port=27017, db_name="jazz")

df_result =, func_request, func_response, calls= 60, period = 60, max_tries = 8)

Clustering Strings

Optimus implements some funciton to cluster Strings. We get graet inspiration from OpenRefine

Here a quote from its site:

"In OpenRefine, clustering refers to the operation of "finding groups of different values that might be alternative representations of the same thing". For example, the two strings "New York" and "new york" are very likely to refer to the same concept and just have capitalization differences. Likewise, "Gödel" and "Godel" probably refer to the same person."

For more informacion see this:


df ="../examples/data/random.csv",header=True, sep=";")
from import keycollision as keyCol
df_kc = keyCol.fingerprint_cluster(df, 'STATE')

keyCol.fingerprint_cluster(df, "STATE").to_json()
df_kc = keyCol.n_gram_fingerprint_cluster(df, "STATE" , 2)

keyCol.n_gram_fingerprint_cluster(df, "STATE" , 2).to_json()

Nearest Neighbor Methods

from import distancecluster as dc
df_dc = dc.levenshtein_matrix(df,"STATE")


df_dc = dc.levenshtein_cluster(df,"STATE")

dc.to_json(df, "STATE")

Machine Learning

Machine Learning is one of the last steps, and the goal for most Data Science WorkFlows.

Apache Spark created a library called MLlib where they coded great algorithms for Machine Learning. Now with the ML library we can take advantage of the Dataframe API and its optimization to create Machine Learning Pipelines easily.

Even though this task is not extremely hard, it is not easy. The way most Machine Learning models work on Spark are not straightforward, and they need lots of feature engineering to work. That's why we created the feature engineering section inside Optimus.

One of the best "tree" models for machine learning is Random Forest. What about creating a RF model with just one line? With Optimus is really easy.

df_cancer = op.load.csv("")
columns = ['diagnosis', 'radius_mean', 'texture_mean', 'perimeter_mean', 'area_mean', 'smoothness_mean',
           'compactness_mean', 'concavity_mean', 'concave points_mean', 'symmetry_mean',

df_predict, rf_model =, columns, "diagnosis")

This will create a DataFrame with the predictions of the Random Forest model.

So lets see the prediction compared with the actual label:["label","prediction"]).table()

The rf_model variable contains the Random Forest model for analysis.

Contributing to Optimus

Contributions go far beyond pull requests and commits. We are very happy to receive any kind of contributions


[Become a backer] and get your image on our README on Github with a link to your site.


[Become a sponsor] and get your image on our README on Github with a link to your site.

Core Team

Argenis Leon



Apache 2.0 ©

Project details

Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

optimuspyspark-2.2.32.tar.gz (65.1 MB view hashes)

Uploaded Source

Built Distribution

optimuspyspark-2.2.32-py3-none-any.whl (65.1 MB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page