Big Data

Collection of datasets being so large, that its difficult to process with the traditional techniques, like RDBMS

The rule of Vs:

Volume - or how much are the data?

Variety - how diverse are different types of data?

Velocity - at what speed are the new data generated?

(Veracity - how accurate is the data?)

Big Data - Tools and Techniques

Apache Hadoop - The Apache™ Hadoop® project develops open-source software for reliable, scalable, distributed computing.

Python in Hadoop® ecosystem:

Use Hadoop® Streaming API and do Map-Reduce with simple python code.

        $HADOOP_HOME/bin/hadoop  jar $HADOOP_HOME/hadoop-streaming.jar \
        -input myInputDirs \
        -output myOutputDir \
        -mapper myPythonScript.py \
        -file myPythonScript.py
      

Machine Learning

Machine learning is a field of computer science that uses statistical techniques to give computer systems the ability to "learn" (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed

Machine Learning

Coined in 1959 by Arthur Samuel

Algorithm that can learn from data and predict new data states

Supervised and Unsupervised Learning

Interweaves the advantages in computational statistics, data mining, big data, neural networks, logic programming and rule-based computing

Machine_learning with Python

numpy

pandas

Tensorflow

Natural Language Processing (NLP)

Conceptualised in the 1950s

ELIZA by by Joseph Weizenbaum

MARGIE by Schank et all., 1975

Evolves rapidly with evolution of Machine Learning and Deep Learning

Google Translate

Inteligent Chat Bots

NLP with Python

NLTK

Stanford CoreNLP

NoSQL Data Stores

Tries to solve the problems of storing and manipulating data, which can not be structured efficiently into tables

Variety of types, most popular:

Key-value pair

Document-oriented

Graph Database

Python and NoSQL stores

pickleDB for key-value pairs databases.

PyMongo for MongoDB document-oriented databases

Neo4j Python Driver for Python Graph databases.

iPython

IPython - an enhanced Interactive Python, which provides:

A powerful interactive shell

Flexible, embeddable interpreters to load into your own projects

Support for interactive data visualization and use of GUI toolkits.

Easy to use, high performance tools for parallel computing.

A kernel for Jupyter.

iPython install

      #install iPython with pip3
      $ pip3 install ipython

      #install iPython on pipenv
      $ pipenv install ipython
    

Jupyter Notebook

Live Demo on Google Colaboratory

Jupyter Notebook - install

      #install jupyter with pip3
      $ pip3 install jupyter

      #install jupyter on pipenv
      $ pipenv install jupyter
    

numpy overview

numpy is a python package that adds support for:

large, multi-dimensional arrays and matrices

large collection of high-level mathematical functions to operate on these arrays

Written mainly in C langualge

numpy arrays and operations are much faster than Python's list equivalents.

pandas is built on top of numpy

data types

Numpy supports much finer data types than Python

Data types @docs.scipy.org

N-Dimensional Arrays

The main object in numpy is the homogeneous multidimensional array

It is a table of elements (usually numbers), all of the same type, indexed by a tuple of positive integers for each dimension.

In numpy, dimensions are called axes

Number of axes defines the rank of the array

Examples

numpy array examples in Jupyter Notebook:

See as html

Or play online (if you have Google Account), or download the notebook: NDArrays.ipynb

References

Functions and Methods Overview

numpy in PyPi

Overview

pandas is an open source library providing high-performance, easy-to-use data structures and data analysis tools for the Python programming language

pandas is great for data analysis and modeling

pandas combined with the IPython toolkit and other libraries creates environment for doing data analysis in Python, which excels in performance, productivity, and the ability to collaborate.

Install

      #install pandas with pip3
      $ pip3 install pandas

      #install pandas on pipenv
      $ pipenv install pandas

      # install pandas on conda
      conda install pandas
    

Pandas Data Structures

The two primary data structures in pandas are Series and DataFrame.

Pandas Series Object

A Pandas Series is a one-dimensional array of indexed data, based on the NumPy ndarray.

But a Pandas Series object wraps both a sequence of values and a sequence of indices

      import pandas as pd

      ds = pd.Series([1,2,3,4])
      print(ds)
    

      0    1
      1    2
      2    3
      3    4
      dtype: int64
    

Create Series with Explicit Indexing

The explicit index definition gives the Series object additional capabilities compared to numpy arrays

i.e. the index need not to be an integer, but can consist of values of any desired type

      ds = pd.Series([1,2,3,4], index=['a', 'b', 'c', 'd'])
      print(ds)
    

      a    1
      b    2
      c    3
      d    4
      dtype: int6
    

Create Series from dictionary

By default, a Series index is drawn from the sorted keys of the dict.

      ds = pd.Series({
        "d":4,
        "a":1,
        "c":3,
        "b":2,
        "e":5
      })
      print(ds)
    

      a    1
      b    2
      c    3
      d    4
      e    5
      dtype: int64
    

Series Indexing

You can use a single index value, or a list of indexes, or slicing

        ##get index object:
        print(ds.index)
        #Index(['apples', 'bananas', 'oranges'], dtype='object')

        ## numerical or keyword indexes
        print(ds["a"])
        print(ds[0])

        ## indexes as list:
        print(ds[['a', 'c', 'e']])
        #a    1
        #c    3
        #e    5
        #dtype: int6
      

Series slicing

the Series also supports array-style operations such as slicing

      ## slicing
      print(ds["a":"d"])
      #a    1
      #b    2
      #c    3
      #d    4
      #dtype: int64
    

Altering index in place

A Series’s index can be altered in place by assignment

      ds = pd.Series([1,2,3,4,5],index=["a","b","c","d","e"])

      ds.index = ["A","B","C","D","E"]
      print(ds)
      #A    1
      #B    2
      #C    3
      #D    4
      #E    5
      #dtype: int64
    

NumPy operations on Series

      ds = pd.Series([1,2,3,4,5],index=["a","b","c","d","e"])

      ## filtering by value
      ds[ds>2]
      #c    3
      #d    4
      #e    5
      #dtype: int64

      ## multiplication
      ds*2
      #a     2
      #b     4
      #c     6
      #d     8
      #e    10
      #dtype: int64
    

Dictionary like operation on Series

      ds = pd.Series([1,2,3,4,5],index=["a","b","c","d","e"])

      "a" in ds
      #True

      "f" in ds
      #False
    

Missing Data

Missing data can appear when we transform or make some operations on Series oject. These data values are marked as NaN (Not A Number) values

      ds1 = pd.Series([1,3], index=["a","c"])
      ds2 = pd.Series([2,3], index=["b","c"])

      print(ds1+ds2)
      #a    NaN
      #b    NaN
      #c    6.0
      #dtype: float64
    

All Examples

Examples as HTML: Series.html

Examples as ipynb: Series.ipynb

Examples as ipynb on Google Colab: Series.ipynb on Google Colab

References

pandas.Series @pandas.pydata.org

Pandas DataFrame Object

a DataFrame is an analogue of a two-dimensional array or table with flexible row and column indices.

You can think of a DataFrame as a sequence of aligned (sharing same index) Series objects

i.e. each column in a DataFrame is represented by a Series Object

Create DataFrame from a single Series object.

A DataFrame is a collection of Series objects, and a single-column DataFrame can be constructed from a single Series.

      # create Series Object:
      prices_ds = pd.Series([1.5, 2, 2.5, 3],
                  index=["apples", "oranges", "bananas", "strawberries"])

      # create DataFrame Object from prices Series:
      prices_df = pd.DataFrame(prices_ds)
      print(prices_df)

      #                0
      #apples        1.5
      #oranges       2.0
      #bananas       2.5
      #strawberries  3
    

Create DataFrame from a single Series object.

We can pass columns names, instead of defaults

      # create DataFrame Object from prices Series:
      prices_df = pd.DataFrame(prices_ds,columns=["prices"])
      print(prices_df)
      #              prices
      #apples           1.5
      #oranges          2.0
      #bananas          2.5
      #strawberries     3.0
    

Create DataFrame from a dictionary of Series object.

A DataFrame can be thought as a dictionary of Series objects, where the dictionary keys represent the columns names.

      prices_ds = pd.Series([1.5, 2, 2.5, 3],
                  index=["apples", "oranges", "bananas", "strawberries"])

      suppliers_ds = pd.Series(["supplier1", "supplier2", "supplier4", "supplier3"],
                     index=["apples", "oranges", "bananas", "strawberries"])

      fruits_df = pd.DataFrame({
        "prices": prices_ds,
        "suppliers": suppliers_ds
      })
      print(fruits_df)
      #              prices  suppliers
      #apples           1.5  supplier1
      #oranges          2.0  supplier2
      #bananas          2.5  supplier4
      #strawberries     3.0  supplier3
    

All Examples

Examples as HTML: DataFrameOverview.html

Examples as ipynb: DataFrameOverview.ipynb

More on pandas.DataFrame() class

class pandas.DataFrame() @pandas.pydata.org

note

In next slides, the term csv is used in its broader meaning - i.e. it will be used for tab/semicolon/whatsoever separated values

the Dataset

In next examples the csv files used are taken from IMDb data files available for download

For test purposes, a shorten versions of the files, containing the first 500 records from each, are available:

title.basics_sample_500.tsv

name.basics_sample_500.tsv

read_csv

read_csv is the preferred method for loading csv data into a DataFrame object

It provides a flourish amount of properties and methods for fine-tune loading, cleaning and manipulation of csv files.

Examples

examples as HTML: imdb.html

example as ipynb: imdb.ipynb

More on pandas.read_csv

pandas.read_csv @pandas.pydata.org

Overview

merge method in pandas is analogues to SQL join operation!

merge is the entry point for all standard database join operations between DataFrame objects:

The related DataFrame.join method, uses merge internally for the index-on-index (by default) and column(s)-on-index join.

Syntax

      pd.merge(left, right, how='inner', on=None, left_on=None, right_on=None,
               left_index=False, right_index=False, sort=True,
               suffixes=('_x', '_y'), copy=True, indicator=False,
               validate=None)
    

the datasets

developers.csv and languages.csv

      did;  dname
      1;    Ivan
      2;    Asen
      3;    Maria
      4;    Stoyan
      5;    Aleks
      6;    Svetlin
    

      did;  language
      2;    "C++"
      3;    "Python"
      3;    "R"
      6;    "Java"
    

Inner Join

Return only the rows in which the left table have matching keys in the right table

      dev_langs_inner = pd.merge(devs,langs,on="did",how='inner')
      print(dev_langs_inner)
    

      did dname language
      0 2 Asen  "C++"
      1 3 Maria "Python"
      2 3 Maria "R"
      3 6 Svetlin "Java"
    

Output Join

Returns all rows from both tables, join records from the left which have matching keys in the right table.

      dev_langs_outer = pd.merge(devs,langs,on="did",how='outer')
      dev_langs_outer
    

      did dname language
      0 1 Ivan  NaN
      1 2 Asen  "C++"
      2 3 Maria "Python"
      3 3 Maria "R"
      4 4 Stoyan  NaN
      5 5 Aleks NaN
      6 6 Svetlin "Java"
    

Left outer join

Return all rows from the left table, and any rows with matching keys from the right table.

      dev_langs_left_outer = pd.merge(devs,langs,on="did",how='left')
      dev_langs_left_outer
    

      did dname language
      0 1 Ivan  NaN
      1 2 Asen  "C++"
      2 3 Maria "Python"
      3 3 Maria "R"
      4 4 Stoyan  NaN
      5 5 Aleks NaN
      6 6 Svetlin "Java"
    

Right outer join

Return all rows from the right table, and any rows with matching keys from the right table.

      dev_langs_right_outer = pd.merge(devs,langs,on="did",how='right')
      dev_langs_right_outer
    

      did dname language
      0 2 Asen  "C++"
      1 3 Maria "Python"
      2 3 Maria "R"
      3 6 Svetlin "Java"
    

All Examples

Examples as HTML: DataFramesMerge.html

Examples as ipynb: DataFramesMerge.ipynb

resources

merge @pandas.pydata.org

Merge, join, and concatenate @pandas.pydata.org

Comparison with SQL @pandas.pydata.org

Overview

Join columns with other DataFrame either on index or on a key column

By default, join() will join the DataFrames on their indices

Efficiently Join multiple DataFrame objects by index at once by passing a list

All Examples

Examples as HTML: DataFramesJoin.html

Examples as ipynb: DataFramesJoin.ipynb

resources