AI/ML Engineer - Guía de Entorno

Guía completa para desarrollo de Inteligencia Artificial y Machine Learning con Claude Code.

Capacidades de Claude Code

Capacidad	Herramientas
Entrenar modelos	Python, PyTorch, TensorFlow, JAX
Jupyter Notebooks	JupyterLab, VS Code Notebooks
MLOps	MLflow, Kubeflow, DVC
LLM Development	LangChain, LlamaIndex, Hugging Face
Computer Vision	OpenCV, YOLO, Detectron2
NLP	spaCy, NLTK, Transformers
Data Processing	Pandas, Polars, Dask
GPU Computing	CUDA, cuDNN, ROCm

---

Python ML Stack

Instalación Base

# Python 3.11+ (compatibilidad ML)
# Windows
winget install Python.Python.3.11

# macOS
brew install python@3.11

# Verificar
python --version
pip --version

Entornos Virtuales (Conda recomendado para ML)

# Miniconda (recomendado)
# Windows
winget install Anaconda.Miniconda3

# macOS
brew install miniconda

# Linux
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

# Crear entorno
conda create -n ml python=3.11
conda activate ml

# O con mamba (más rápido)
conda install mamba -c conda-forge
mamba create -n ml python=3.11

Frameworks de Deep Learning

# PyTorch (recomendado)
# CPU
pip install torch torchvision torchaudio

# CUDA 12.1 (NVIDIA GPU)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

# Apple Silicon (MPS)
pip install torch torchvision torchaudio  # MPS incluido automáticamente

# TensorFlow
pip install tensorflow

# Con GPU
pip install tensorflow[and-cuda]

# JAX (Google)
pip install jax jaxlib

# Con GPU
pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

Librerías Esenciales

# Data Science
pip install numpy pandas polars scikit-learn scipy matplotlib seaborn

# Deep Learning extras
pip install transformers datasets accelerate  # Hugging Face
pip install lightning  # PyTorch Lightning
pip install keras  # High-level API

# Computer Vision
pip install opencv-python pillow albumentations
pip install ultralytics  # YOLO
pip install detectron2  # Facebook

# NLP
pip install spacy nltk gensim
python -m spacy download es_core_news_lg
python -m spacy download en_core_web_lg

# LLM Development
pip install langchain langchain-community langchain-openai
pip install llama-index
pip install openai anthropic

# MLOps
pip install mlflow wandb tensorboard
pip install dvc  # Data Version Control
pip install optuna  # Hyperparameter tuning

# Notebooks
pip install jupyterlab notebook ipywidgets

---

GPU Setup

NVIDIA CUDA

# Windows
winget install Nvidia.CUDA

# Linux (Ubuntu)
# Seguir guía oficial: https://developer.nvidia.com/cuda-downloads

# Verificar instalación
nvidia-smi
nvcc --version

# Python - verificar GPU disponible
python -c "import torch; print(torch.cuda.is_available())"
python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Apple Silicon (M1/M2/M3)

# MPS (Metal Performance Shaders) incluido en PyTorch
python -c "import torch; print(torch.backends.mps.is_available())"

# TensorFlow con Metal
pip install tensorflow-metal

AMD ROCm

# Linux only
# https://rocm.docs.amd.com/

# PyTorch con ROCm
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.6

---

Jupyter Environment

JupyterLab

# Instalar
pip install jupyterlab

# Extensiones útiles
pip install jupyterlab-git
pip install jupyterlab-lsp python-lsp-server

# Ejecutar
jupyter lab

# Con puerto específico
jupyter lab --port 8889 --no-browser

VS Code Notebooks

# Extensiones
code --install-extension ms-toolsai.jupyter
code --install-extension ms-toolsai.vscode-jupyter-cell-tags
code --install-extension ms-toolsai.vscode-jupyter-slideshow
code --install-extension ms-python.python

---

MLOps Tools

MLflow

pip install mlflow

# Servidor local
mlflow server --host 0.0.0.0 --port 5000

# UI: http://localhost:5000

# En código
import mlflow
mlflow.set_tracking_uri("http://localhost:5000")
mlflow.set_experiment("my-experiment")

with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_metric("accuracy", 0.95)
    mlflow.pytorch.log_model(model, "model")

Weights & Biases

pip install wandb

# Login
wandb login

# En código
import wandb
wandb.init(project="my-project")
wandb.log({"loss": 0.5, "accuracy": 0.95})

DVC (Data Version Control)

pip install dvc dvc-s3  # o dvc-gdrive, dvc-azure

# Inicializar
dvc init

# Trackear datos
dvc add data/dataset.csv

# Remote storage
dvc remote add -d myremote s3://mybucket/dvc
dvc push
dvc pull

---

LLM Development

LangChain

pip install langchain langchain-openai langchain-anthropic langchain-community

# Ejemplo
from langchain_anthropic import ChatAnthropic
from langchain.prompts import ChatPromptTemplate

llm = ChatAnthropic(model="claude-3-sonnet-20240229")
prompt = ChatPromptTemplate.from_template("Explica {topic} en términos simples")
chain = prompt | llm
response = chain.invoke({"topic": "machine learning"})

Hugging Face

pip install transformers datasets accelerate

# Login para modelos privados
huggingface-cli login

# Descargar modelo
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")

Ollama (LLMs locales)

# Instalar
# Windows/macOS: https://ollama.ai/download
# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Ejecutar modelo
ollama run llama2
ollama run codellama
ollama run mistral

# API local
curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
  "prompt": "Hello"
}'

# Con LangChain
from langchain_community.llms import Ollama
llm = Ollama(model="llama2")

---

Docker para ML

Dockerfile

FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime

WORKDIR /app

# Dependencias del sistema
RUN apt-get update && apt-get install -y \
    git \
    && rm -rf /var/lib/apt/lists/*

# Dependencias Python
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

CMD ["python", "train.py"]

Docker Compose con GPU

version: '3.8'

services:
  training:
    build: .
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    volumes:
      - ./data:/app/data
      - ./models:/app/models
    environment:
      - WANDB_API_KEY=${WANDB_API_KEY}

  mlflow:
    image: ghcr.io/mlflow/mlflow:latest
    ports:
      - "5000:5000"
    command: mlflow server --host 0.0.0.0
    volumes:
      - mlflow_data:/mlflow

  jupyter:
    build: .
    ports:
      - "8888:8888"
    command: jupyter lab --ip=0.0.0.0 --allow-root --no-browser
    volumes:
      - .:/app
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  mlflow_data:

---

Comandos que Claude Code Ejecutará

# Entorno
conda create -n ml python=3.11
conda activate ml
pip install -r requirements.txt

# Jupyter
jupyter lab
jupyter nbconvert --to script notebook.ipynb

# Training
python train.py
python train.py --epochs 100 --lr 0.001
accelerate launch train.py  # Multi-GPU

# MLflow
mlflow run . -P epochs=100
mlflow models serve -m runs:/<run_id>/model -p 5001

# DVC
dvc add data/
dvc push
dvc pull

# Hugging Face
huggingface-cli download meta-llama/Llama-2-7b-hf
python -m transformers.onnx --model=bert-base-uncased onnx/

# Ollama
ollama pull llama2
ollama run codellama "Write a Python function to sort a list"

# GPU
nvidia-smi
watch -n 1 nvidia-smi  # Monitor en tiempo real

---

Verificación del Entorno

#!/bin/bash
echo "=== Verificación Entorno AI/ML ==="

echo -e "\n--- Python ---"
python --version
which python

echo -e "\n--- Conda ---"
conda --version 2>/dev/null || echo "Conda no instalado"

echo -e "\n--- PyTorch ---"
python -c "import torch; print(f'PyTorch: {torch.__version__}')" 2>/dev/null || echo "PyTorch no instalado"
python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')" 2>/dev/null
python -c "import torch; print(f'MPS available: {torch.backends.mps.is_available()}')" 2>/dev/null

echo -e "\n--- TensorFlow ---"
python -c "import tensorflow as tf; print(f'TensorFlow: {tf.__version__}')" 2>/dev/null || echo "TensorFlow no instalado"
python -c "import tensorflow as tf; print(f'GPUs: {tf.config.list_physical_devices(\"GPU\")}')" 2>/dev/null

echo -e "\n--- GPU ---"
nvidia-smi 2>/dev/null || echo "nvidia-smi no disponible"

echo -e "\n--- MLOps ---"
mlflow --version 2>/dev/null || echo "MLflow no instalado"
dvc --version 2>/dev/null || echo "DVC no instalado"

echo -e "\n--- LLM Tools ---"
ollama --version 2>/dev/null || echo "Ollama no instalado"

echo -e "\n=== Verificación Completa ==="

---

Recursos

• PyTorch Tutorials
• Hugging Face Course
• LangChain Docs
• MLflow Docs
• Fast.ai