Mahmoud Abas, Full-Stack Software Engineer & AI

Software Construction - ML Week01 Lab

In the print dialog, choose "Save as PDF" as the destination.

ML Week 1 Lab

Fundamentals of Machine Learning, then hands-on NumPy, Pandas and Matplotlib

Term Plan Overview

Fundamentals of ML
Linear & Logistic Regression
K-Nearest Neighbours
Model Evaluation & Selection
Support Vector Machines
Decision Trees, Naive Bayes, Neural Networks

What Is Machine Learning?

Machine Learning: The study of algorithms that can learn by example.

ML algorithms generalize from existing examples of a task, for example after seeing labeled training images, an image classifier can label new, unseen images accurately
Brings together statistics, computer science, and more
Algorithms learn rules from labelled training data, and those rules must generalize to new examples not seen during training
Pure ML studies algorithm behavior in general; applied ML studies behavior on specific real-life datasets

Types of ML Problems

Supervised Learning

Unsupervised Learning

Reinforcement / Semi-Supervised

Supervised learning: learn to predict target values from labelled data. Classification (discrete classes) covers fraud detection, spam detection, diagnostics, image classification; regression (continuous values) covers risk assessment, score prediction
Unsupervised learning: find structure in unlabeled data. Clustering finds groups of similar instances; outlier detection finds unusual patterns
Reinforcement / semi-supervised learning: algorithms that use mostly unlabeled data plus a little labeled data

Why Python for Machine Learning

Readable, maintainable code: clean syntax reduces time spent on upkeep
Runs on major platforms; interpreted, so the same code runs everywhere without recompilation
Supports multiple paradigms: object-oriented, structured, functional/aspect-oriented; dynamic typing and automatic memory management
Powerful open-source libraries for data analysis and visualization: NumPy, Pandas, Matplotlib
Anaconda bundles Python plus 720+ open-source packages as a free environment/package manager

Python Modules and the ML Toolchain

Module: A single Python file imported with import module_name (for example math, random, os, datetime, csv).

Package: A directory hierarchy of modules, imported with dot notation (for example Numpy, Pandas, Matplotlib).

python

import numpy as np          # fundamental n-dimensional arrays
import pandas as pd         # DataFrame: tabular data manipulation
import matplotlib.pyplot as plt   # visualization
import scipy as sp          # scientific computing, sparse matrices
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier

The ML toolchain, one import per library.

NumPy: Creating Arrays

python

arr = np.array([0, 1, 2, 3, 4])
np.arange(0, 10)          # array([0, 1, ..., 9])
np.arange(0, 10, 3)       # array([0, 3, 6, 9])
np.zeros((5, 5))          # 5x5 array of 0.
np.ones((2, 4))           # 2x4 array of 1.
np.random.randint(0, 100, (3, 3))   # random 3x3 int array
np.linspace(0, 10, 6)     # 6 evenly spaced values from 0 to 10

NumPy: Operations and Indexing

python

arr2.max()       # 95
arr2.min()       # 9
arr2.mean()      # 60.8
arr2.argmin()    # index of the minimum
arr2.argmax()    # index of the maximum
arr2.reshape(2, 5)   # reshape a flat array into 2x5

mat = np.arange(0, 100).reshape(10, 10)
mat[row, col]     # single element
mat[:, 0]         # entire first column
mat[0:3, 0:3]     # top-left 3x3 slice

Pandas: Series and DataFrames

pandas.Series: A labeled, one-dimensional array (like a smarter list with an index).

pandas.DataFrame: A labeled, two-dimensional table (like a spreadsheet).

python

step_counts = pd.Series(step_data, name='steps')
step_counts.index = pd.date_range('20240329', periods=6)
print(step_counts['2024-04-01'])   # access by label, like a dict
print(step_counts[3])              # access by position, like an array

activity_df = pd.DataFrame(joined_data,
                            index=pd.date_range('20240329', periods=6),
                            columns=['Walking', 'Cycling'])

Pandas: Selecting Data with loc and iloc

.loc: Label-based selection; includes the last element of the requested range; also accepts boolean masks.

.iloc: Integer position-based selection, like array indexing.

python

data = pd.read_csv('Iris_Data.csv')
data.head()                          # first 5 rows

print(data.loc[2, ['sepal_length']]) # row labeled 2, one column
print(data.iloc[:5, :])              # first 5 rows, all columns
print(data.iloc[:, 0])               # first column, all rows

Pandas: Filtering, New Columns, Grouping

python

# Filter rows with a boolean condition
print(data.loc[(data.sepal_width > 3.5)])

# Derive a new column from existing ones
data['sepal_area'] = data.sepal_length * data.sepal_width

# Count rows per category
group_sizes = data.groupby('species').size()

Pandas: Descriptive Statistics

Mean: The average value.

Median: The midpoint value.

Mode: The most common value.

Standard deviation: How spread out values are around the mean (low means tightly clustered, high means widely spread); it is the square root of the variance.

Standard error of the mean (SEM): How far the sample mean is likely to be from the true population mean; always smaller than the standard deviation.

python

print(data.petal_length.mean())
print(data.petal_length.median())
print(data.petal_length.std())
print(data.petal_length.var())
print(data.petal_length.sem())

Matplotlib: Basic Plotting

python

ax = plt.axes()
ax.hist(data.petal_length, bins=25)
ax.set(xlabel='Petal Length (cm)', ylabel='Frequency',
       title='Distribution of Petal Lengths')

plt.plot(data.sepal_length, data.sepal_width, marker='o')
plt.title("Distribution of Petal Lengths")
plt.xlabel("Petal Length (cm)")
plt.ylabel("sepal_width")

Key Takeaways

ML learns generalizable rules from labelled training examples; applied ML evaluates those algorithms on real datasets
Supervised learning covers classification (discrete) and regression (continuous); unsupervised learning covers clustering and outlier detection
Python's readability, portability, and library ecosystem (NumPy, Pandas, Matplotlib, scikit-learn, SciPy) make it the standard ML language
NumPy provides fast n-dimensional arrays with creation, reshaping, and indexing/slicing operations
Pandas Series/DataFrame give labeled, spreadsheet-like data with .loc (label-based) and .iloc (position-based) selection, filtering, and grouping
Matplotlib turns DataFrame columns into histograms and plots for quick visual exploration

Open this lesson

Mahmoud AbasML Week01 Lab

import numpy as np # fundamental n-dimensional arrays import pandas as pd # DataFrame: tabular data manipulation import matplotlib.pyplot as plt # visualization import scipy as sp # scientific computing, sparse matrices from sklearn.model_selection import train_test_split from sklearn.tree import DecisionTreeClassifier

arr = np.array([0, 1, 2, 3, 4]) np.arange(0, 10) # array([0, 1, ..., 9]) np.arange(0, 10, 3) # array([0, 3, 6, 9]) np.zeros((5, 5)) # 5x5 array of 0. np.ones((2, 4)) # 2x4 array of 1. np.random.randint(0, 100, (3, 3)) # random 3x3 int array np.linspace(0, 10, 6) # 6 evenly spaced values from 0 to 10

arr2.max() # 95 arr2.min() # 9 arr2.mean() # 60.8 arr2.argmin() # index of the minimum arr2.argmax() # index of the maximum arr2.reshape(2, 5) # reshape a flat array into 2x5 mat = np.arange(0, 100).reshape(10, 10) mat[row, col] # single element mat[:, 0] # entire first column mat[0:3, 0:3] # top-left 3x3 slice

step_counts = pd.Series(step_data, name='steps') step_counts.index = pd.date_range('20240329', periods=6) print(step_counts['2024-04-01']) # access by label, like a dict print(step_counts[3]) # access by position, like an array activity_df = pd.DataFrame(joined_data, index=pd.date_range('20240329', periods=6), columns=['Walking', 'Cycling'])

data = pd.read_csv('Iris_Data.csv') data.head() # first 5 rows print(data.loc[2, ['sepal_length']]) # row labeled 2, one column print(data.iloc[:5, :]) # first 5 rows, all columns print(data.iloc[:, 0]) # first column, all rows

# Filter rows with a boolean condition print(data.loc[(data.sepal_width > 3.5)]) # Derive a new column from existing ones data['sepal_area'] = data.sepal_length * data.sepal_width # Count rows per category group_sizes = data.groupby('species').size()

ax = plt.axes() ax.hist(data.petal_length, bins=25) ax.set(xlabel='Petal Length (cm)', ylabel='Frequency', title='Distribution of Petal Lengths') plt.plot(data.sepal_length, data.sepal_width, marker='o') plt.title("Distribution of Petal Lengths") plt.xlabel("Petal Length (cm)") plt.ylabel("sepal_width")