Dvc

DVC (Data Version Control)

Installation

# Install DVC with cloud storage support
pip install dvc[s3]    # For AWS S3
pip install dvc[gs]    # For Google Cloud Storage
pip install dvc[azure] # For Azure Blob Storage
pip install dvc[all]   # All remotes

# Initialize DVC in a Git repo
cd my-ml-project
git init
dvc init

Track Data Files

# track_data.sh — Add large files to DVC tracking instead of Git
# Add a large dataset
dvc add data/training_images/
dvc add data/dataset.csv

# DVC creates .dvc files (small pointers) — commit those to Git
git add data/training_images.dvc data/dataset.csv.dvc .gitignore
git commit -m "Track training data with DVC"

Configure Remote Storage

# setup_remote.sh — Configure where DVC stores actual file contents
# S3
dvc remote add -d myremote s3://my-bucket/dvc-storage

# Google Cloud Storage
dvc remote add -d myremote gs://my-bucket/dvc-storage

# Local / network path
dvc remote add -d myremote /mnt/shared/dvc-storage

# Push data to remote
dvc push

# Pull data from remote (on another machine or after cloning)
dvc pull

Build Reproducible Pipelines

# dvc.yaml — Define ML pipeline stages with dependencies and outputs
stages:
  prepare:
    cmd: python src/prepare.py
    deps:
      - src/prepare.py
      - data/raw/
    outs:
      - data/processed/

  train:
    cmd: python src/train.py
    deps:
      - src/train.py
      - data/processed/
    params:
      - train.epochs
      - train.learning_rate
      - train.batch_size
    outs:
      - models/model.pkl
    metrics:
      - metrics/train.json:
          cache: false

  evaluate:
    cmd: python src/evaluate.py
    deps:
      - src/evaluate.py
      - models/model.pkl
      - data/processed/
    metrics:
      - metrics/eval.json:
          cache: false
    plots:
      - metrics/confusion_matrix.csv:
          x: predicted
          y: actual

# params.yaml — Pipeline parameters (tracked by DVC)
train:
  epochs: 50
  learning_rate: 0.001
  batch_size: 32

# Run the entire pipeline (only re-runs changed stages)
dvc repro

# Run a specific stage
dvc repro train

Experiment Tracking

# experiments.sh — Run and compare ML experiments
# Run an experiment with modified parameters
dvc exp run --set-param train.learning_rate=0.01

# Run multiple experiments in parallel
dvc exp run --set-param train.learning_rate=0.001 --queue
dvc exp run --set-param train.learning_rate=0.01 --queue
dvc exp run --set-param train.learning_rate=0.1 --queue
dvc queue start --jobs 3

# Compare experiments
dvc exp show
dvc exp diff

# Apply a successful experiment to workspace
dvc exp apply exp-abc123

# Push experiment to Git branch
dvc exp push origin exp-abc123

Metrics and Plots

# train.py — Training script that outputs DVC-tracked metrics
import json
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, f1_score
import yaml
import pickle

# Load params
with open("params.yaml") as f:
    params = yaml.safe_load(f)["train"]

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

model = RandomForestClassifier(n_estimators=params["epochs"])
model.fit(X_train, y_train)

preds = model.predict(X_test)
metrics = {
    "accuracy": accuracy_score(y_test, preds),
    "f1_score": f1_score(y_test, preds, average="weighted"),
}

with open("metrics/train.json", "w") as f:
    json.dump(metrics, f, indent=2)

with open("models/model.pkl", "wb") as f:
    pickle.dump(model, f)

# View metrics across experiments
dvc metrics show
dvc metrics diff

# Generate plots
dvc plots show metrics/confusion_matrix.csv
dvc plots diff  # Compare plots between experiments

Data Access Without Cloning

# Access tracked files from any DVC repo without full clone
dvc get https://github.com/org/ml-repo data/processed/dataset.csv
dvc import https://github.com/org/ml-repo models/model.pkl

# dvc_api.py — Access DVC-tracked files programmatically
import dvc.api

# Read a file from a DVC repo
with dvc.api.open("data/dataset.csv", repo="https://github.com/org/ml-repo") as f:
    import pandas as pd
    df = pd.read_csv(f)

# Get the URL of a tracked file
url = dvc.api.get_url("models/model.pkl", repo="https://github.com/org/ml-repo")

Key Concepts

• .dvc files: Small pointer files committed to Git that reference large data in remote storage
• dvc repro: Reproduce pipelines — only re-runs stages with changed dependencies
• Experiments: Branch-free experiment tracking — run, compare, and apply results
• Params: YAML parameter files tracked by DVC for reproducible configurations
• Metrics: JSON/YAML metrics files with built-in comparison tools
• Remote storage: S3, GCS, Azure, SSH, HDFS — data stays where you want it