A pandas DataFrame is a two-dimensional, table-like structure for organizing tabular data as an ordered collection of labeled columns and rows. It can keep numeric, ordinal, categorical, date, string, and boolean values together while preserving their appropriate data types. Row and column indexes make records and features addressable by label or by zero-based position. section 2.1.3
Use a DataFrame as the working table for data exploration and preparation. It supports loading data from multiple sources; selecting rows and columns; handling missing values; adding, renaming, computing, grouping, and aggregating features; and pivoting, reshaping, and visualizing results. section 2.1.3
A DataFrame is useful when tabular work needs to stay inspectable and incremental:
- Build a small dataset directly from a Python dictionary, then assign meaningful row labels and feature names. section 2.1.3
- Retrieve one record with either positional indexing or its row label. section 2.1.3
- Join product tables on a shared identifier with pandas merge, keeping intermediate transformation stages available for checking. section 2.3.1
- Prepare heterogeneous features for exploratory data analysis or machine-learning input while retaining data types and ordering. section 2.3.1 section 2.5
A labeled customer-style table
Create four rows for Bob, Alice, Charlie, and Emily with gender, age, education level, and income columns. Converting the dictionary to a DataFrame produces a labeled table; Alice’s row can then be selected as position 1 with iloc or by name with loc. The contrast matters: positions depend on row order, whereas labels identify the intended record. section 2.1.3
DataFrames overlap with relational-table workflows: selection, filtering, aggregation, ordering, and processing follow familiar SQL-like principles, while a DataFrame index can play a role similar to a primary key. They provide close control over transformation steps and intermediate results, but that control also makes mistakes easier to introduce. For data science, this is valuable when several sources must be combined, checked, transformed, and plotted before modeling. section 2.3.1
In a containerized training workflow, separate CSV files matching train, validation, and test URI patterns can be read and concatenated into DataFrames. The resulting tables let the training code consume dataset subsets after cloud-storage ingestion. section 11.3.5
Listing 2.1 Creating a simple tabular dataset
import pandas as pd
data = {'gender': ['male', 'female', 'male', 'female'],
'age': [25, 30, 27, 29],
'education_level': ['Bachelor', 'Master', 'Bachelor', 'PhD'],
'income': [50000, 60000, 55000, 70000]} #1
index = [ 'Bob ', 'Alice', 'Charlie', 'Emily'] #2
df = pd.DataFrame(data, index=index) #3
print(df)
print(df.iloc[1]) #4
print(df.loc['Alice']) #5
This listing shows the complete minimal pattern: define heterogeneous columns, supply row labels, construct a DataFrame, and access the same row by position or label. section 2.1.3
