Software Toolkit

Software Toolkit#

The examples in this book use a small collection of Python packages. You do not need to master them before starting. It is enough to know which job each package performs and to recognize the main data structures as they move through a modeling workflow. However, reading their documentation will be beneficial in the long run.

Jupyter notebooks and Google Colab#

A Jupyter notebook mixes explanatory text, executable Python, figures, and saved results. Run cells from top to bottom because later cells usually depend on variables created earlier. Google Colab runs the same kind of notebook in a temporary cloud environment, so downloaded data and installed extras may need to be recreated when the runtime restarts.

Package map#

Package

Typical import

Role in this book

NumPy

import numpy as np

Numerical arrays and vectorized calculations

pandas

import pandas as pd

Dataframes, timestamps, and tabular data cleaning

Matplotlib

import matplotlib.pyplot as plt

Scientific Figures

scikit-learn

from sklearn import ...

Data splits, preprocessing, baselines, metrics, and diagnostic displays

PyTorch

import torch

Canonical neural betwork models, and automatic differentiation

Keras 3

import keras

A higher-level model API using the same PyTorch backend

XGBoost

import xgboost as xgb

Gradient-boosted tree models for tabular and flattened inputs

Optuna / KerasTuner

import optuna / import keras_tuner

Validation-based hyperparameter searches

SHAP

import shap

Model-behavior diagnostics that require careful interpretation

Other useful packages#

The following packages are also particularly useful to use in future exampels or experimentation.

Package

Typical import

When it may be useful

tslearn

import tslearn

Time-series distances, clustering, classification, and related learning tools

sktime

import sktime

A unified interface for forecasting, classification, transformation, and other time-series tasks

Seaborn

import seaborn as sns

Higher-level statistical graphics built on Matplotlib

LightGBM

import lightgbm as lgb

Efficient gradient-boosted tree models, especially for tabular data, similar to XGBoost

imbalanced-learn

import imblearn

Resampling methods, pipelines, and metrics for imbalanced classification problems

statsmodels

import statsmodels.api as sm

Statistical models, hypothesis tests, classical time-series methods, and detailed inference summaries

Three common data representations#

A pandas.DataFrame keeps column names and timestamps, which is helpful while auditing and cleaning scientific tables. NumPy arrays provide compact numerical matrices for preprocessing and many classical models.

import pandas as pd
import torch

frame = pd.DataFrame({"speed": [400.0, 525.0], "bz": [-2.0, 4.0]})
array = frame[["speed", "bz"]].to_numpy(dtype="float32")
tensor = torch.from_numpy(array)

PyTorch is the canonical path#

PyTorch exposes the important steps explicitly: create tensors, define a model, calculate a loss, backpropagate, update parameters, and evaluate with gradients disabled. Keras 3 offers a shorter compile() and fit() workflow while still using PyTorch underneath. The backend must be selected before importing Keras:

import os

os.environ["KERAS_BACKEND"] = "torch"
import keras

assert keras.backend.backend() == "torch"

The Keras notebooks are alternative implementations, not a second modeling method. Their data boundaries, architecture intent, metrics, and scientific interpretation are almost identical to the PyTorch workflow.