The goal

You want an environment that:

• is reproducible (same versions)
• works for both notebooks and scripts
• supports classical ML, plus deep learning later

Recommended baseline stack

Core

• numpynumpy, pandaspandas (arrays + dataframes)
• matplotlibmatplotlib, seabornseaborn (visualization)
• scikit-learnscikit-learn (classic ML: regression, classification, clustering)

Deep learning

Pick at least one:

• torchtorch (PyTorch)
• tensorflowtensorflow (TensorFlow / Keras)

Optional but helpful

• jupyterlabjupyterlab (notebooks)
• ipykernelipykernel (bind env to Jupyter)
• joblibjoblib (save models)

Environment options

Option A — venv (simple)

• good for most projects
• built into Python

Option B — conda (popular for ML)

• excellent for scientific stacks
• can manage non-Python dependencies

Either is fine. The key is: don’t install everything globally.

Versioning and reproducibility

In real ML work, reproducibility matters.

• pin versions (requirements.txtrequirements.txt or pyproject.tomlpyproject.toml)
• record random seeds where needed
• log dataset versions

CPU vs GPU (what to know)

• Scikit-learn is mostly CPU-based.
• Deep learning frameworks can use GPU.

If you don’t have a GPU, you can still learn everything.

Quick sanity check script

Use this tiny script to confirm the stack imports.

import numpy as np
import pandas as pd
import sklearn
 
print("numpy:", np.__version__)
print("pandas:", pd.__version__)
print("sklearn:", sklearn.__version__)
 
try:
    import torch
    print("torch:", torch.__version__)
except Exception as e:
    print("torch: not installed", e)
 
try:
    import tensorflow as tf
    print("tensorflow:", tf.__version__)
except Exception as e:
    print("tensorflow: not installed", e)

import numpy as np
import pandas as pd
import sklearn
 
print("numpy:", np.__version__)
print("pandas:", pd.__version__)
print("sklearn:", sklearn.__version__)
 
try:
    import torch
    print("torch:", torch.__version__)
except Exception as e:
    print("torch: not installed", e)
 
try:
    import tensorflow as tf
    print("tensorflow:", tf.__version__)
except Exception as e:
    print("tensorflow: not installed", e)

Common pitfalls

• Installing both TensorFlow and PyTorch in the same environment can be heavy.
• GPU installs can be OS/driver-specific.

If you want a minimal setup to start Phase 1–7, you can do scikit-learn only and add deep learning libraries later.