Data Science Knowledge Base

Aus MattWiki

This article is a knowledge base with basics for how to start a data science project.

Notable sources:

Python Requirements

  • Python 3 + PIP
 pip install numpy pandas matplotlib seaborn scikit-learn

EDA / Exploratory Data Analysis

See openHPI Jupyter Notebook: EDA-basic-recipe.ipynb

# Load numpy and pandas libraries
import numpy as np
import pandas as pd

# Read data from CSV file into a dataframe
df = pd.read_csv('911.csv')                

# Show informations about columns, and number and data type of their content
print(df.info())

# Show first and last rows and columns of the dataframe
print(df)

# Show first 10 columns of dataframe
print(df.head(10))

# Describe numerical columns of dataframe by showing their min, max, count, mean and other:
print(df.describe())

# Analyze columns of interest, i.e. ZIP code, title and timeStamp:
print(df["zip"].mean())
print(df["zip"].value_counts().head(10))
print(df["zip"].value_counts().tail(10))
print(df["zip"].nunique())
print(df["title"].nunique())
print(df["timeStamp"].min())
print(df["timeStamp"].max())

Finish the exploratory data analysis by writing a management summary containing gained knowledge about the dataset.

Handling Missing Data with Pandas

https://pandas.pydata.org/docs/user_guide/missing_data.html

Location and Dispersion Metrics

Location metrics:

df["*nameOfAColumn*"].mode()
df["*nameOfAColumn*"].mean()
df["*nameOfAColumn*"].median()  

Dispersion metrics:

df["length_in_minutes"].std()
df["length_in_minutes"].quantile(0.75)

We can also calculate the IQR as shown in the video:

print (df["length_in_minutes"].quantile(0.75) - df["length_in_minutes"].quantile(0.25))

Bivariate Analysis

See openHPI Jupyter Notebook: Bivariate_Analsis.ipynb

Introduction

Bivariate analysis can be used for:

  • Exploring the relationship between two variables
  • Comparing groups
  • Testin hypotheses
  • Predicting outcomes

Correlation between two variables is a statistical measure of the strength and direction between them

  • Perfect positive correlation = -1
  • Perfect negative correlation = -1
  • No correlation = 0

Correlation with Scatter Plot

A scatter plot can be used to visualize the relationship between two continous variables.

Correlation using Pandas and Seaborn

import numpy as np
import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns

iris = sns.load_dataset("iris")

print(iris.sample(10))

sns.pairplot(iris)
plt.show()
sns.pairplot(iris, hue="species", diag_kind='hist')
plt.show()

Multivariate Analysis

See openHPI Jupyter Notebook: multivariate-analysis-video.ipynb

Enables for predicting how individual parameters influence the selected parameter, i.e.:

  • How much does the price of a car vary depending of
    • Age
    • KM
    • With or without power windows

Linear Regression

  • Used in Machine Learning
  • Used to determine relationship between independent and dependent variables, which both are continious

Simply Speaking: Linear regression is basically fitting a line to a dataset using least squares method.

See openHPI Jupyter Notebook: Linear_Regression.ipynb

Documentation: https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LinearRegression.html

import numpy as np
import matplotlib.pyplot as plt
from sklearn.linear_model import LinearRegression

# Generate random data with positive correlation
x_val = np.random.rand(100)
y_val = x_val + np.random.random(100)*0.5

# Plot data in scatter plot
plt.scatter(x_val, y_val)

# Reshape into 2D
x = x_val.reshape(-1,1)
y = y_val.reshape(-1,1)

# Create linear regression object
model = LinearRegression()

# Fit model to the data
model.fit(x,y)

# Generate predicted values of y
y_pred = model.predict(x)

# Plot data points and regression line
plt.scatter(x_val, y_val)
plt.plot(x_val, y_pred, color='green')
plt.xlabel('X Values')
plt.ylabel('Y Values')
plt.title('Linear Regression Example')
plt.show()

Decision Trees

See openHPI Jupyter Notebook: Decision_Trees.ipynb

Types of decision trees:

  • Classify things into categories → Classification Tree
  • Predict numeric values → Regression Tree

Logistic Regression

See openHPI Jupyter Notebook: Logistic_Regression_IRIS_video.ipynb