Description
Apply the changes from a specific experiment to your current workspace. This command updates your working directory with the code, parameters, and data from a completed experiment, allowing you to continue working from that state or promote it to a regular Git commit.
This command modifies your workspace. Any uncommitted changes will be overwritten. Commit or stash your current work before applying an experiment.
Usage
dvc exp apply <experiment>
Arguments
The experiment name to apply to your workspace.Can be:
- Auto-generated experiment name (e.g.,
exp-a1b2c)
- Custom experiment name (specified with
-n during dvc exp run)
dvc exp apply exp-44136
dvc exp apply my-experiment
Options
Fail if this command would overwrite conflicting changes in the workspace.The --no-force option is deprecated and will be removed in a future DVC release.
Examples
Apply a successful experiment
# First, view your experiments
dvc exp show
# Apply the best performing experiment
dvc exp apply exp-a1b2c
Changes for experiment 'exp-a1b2c' have been applied to your workspace.
After applying, your workspace now matches the state of the experiment including code, parameters, and outputs.
Apply and commit
# Apply experiment to workspace
dvc exp apply high-accuracy-run
# Review changes
git status
dvc diff
# Commit if satisfied
git add .
git commit -m "Apply high accuracy experiment"
Changes for experiment 'high-accuracy-run' have been applied to your workspace.
On branch main
Changes not staged for commit:
modified: params.yaml
modified: dvc.lock
Committed changes successfully.
Applying an experiment doesn’t automatically create a Git commit. Review the changes first, then commit when ready.
Apply with existing changes
# You have uncommitted changes
git status
On branch main
Changes not staged for commit:
modified: train.py
modified: params.yaml
# Stash current work
git stash
# Apply experiment
dvc exp apply exp-d3e4f
# Review applied changes
git diff
# If needed, restore your previous work
git stash pop
DVC automatically creates a stash at refs/exps/apply/stash before applying. You can revert using:git reset --hard
git stash apply refs/exps/apply/stash
Compare before applying
# First, compare the experiment with current workspace
dvc exp diff HEAD exp-new-model
# If the changes look good, apply them
dvc exp apply exp-new-model
Metric
────────────────────────────────────────────────────────────
Path Metric HEAD exp-new-model diff
────────────────────────────────────────────────────────────
metrics.json accuracy 0.85 0.92 +0.07
metrics.json loss 0.312 0.234 -0.078
Changes for experiment 'exp-new-model' have been applied to your workspace.
Apply and create a branch
# Create a new branch for the experiment
git checkout -b feature/new-model
# Apply experiment
dvc exp apply exp-breakthrough
# Commit changes
git add .
git commit -m "feat: Apply breakthrough model architecture"
# Push to remote
git push -u origin feature/new-model
This workflow is useful for code review: apply the experiment to a branch, then create a pull request.
What Gets Applied
When you apply an experiment, DVC updates:
1. Parameters
params.yaml and other parameter files are updated with experiment values- Custom parameter files specified in
dvc.yaml are also updated
2. Code Changes
- Any code modifications made during the experiment are applied
- This includes changes to training scripts, preprocessing code, etc.
3. DVC Files
dvc.lock is updated with the experiment’s pipeline state- Cached outputs from the experiment become accessible
4. Metrics
- Metric files are updated to reflect experiment results
Large data files tracked by DVC are not copied to your workspace. Instead, DVC updates pointers to the cached data. Use dvc checkout if you need to materialize the data files.
Common Workflows
Experiment Selection Workflow
# 1. Run multiple experiments
dvc exp run --queue -S train.lr=0.1 -n "lr-0.1"
dvc exp run --queue -S train.lr=0.01 -n "lr-0.01"
dvc exp run --queue -S train.lr=0.001 -n "lr-0.001"
dvc exp run --run-all
# 2. Compare results
dvc exp show --sort-by accuracy --sort-order desc
# 3. Apply the best one
dvc exp apply lr-0.001
# 4. Commit to Git
git add .
git commit -m "feat: Apply best learning rate experiment"
Safe Experimentation Workflow
# 1. Save current state
git stash
# 2. Apply and test experiment
dvc exp apply promising-experiment
# 3. Run additional validation
python scripts/validate.py
# 4a. If good, commit
git add .
git commit -m "Apply promising experiment"
# 4b. If not good, revert
git reset --hard HEAD
git stash apply refs/exps/apply/stash
# 1. Apply production-ready experiment
git checkout main
dvc exp apply production-candidate
# 2. Tag the release
git add .
git commit -m "release: Deploy model v2.0"
git tag -a v2.0 -m "Model version 2.0 with 92% accuracy"
# 3. Push to production
git push origin main --tags
Reverting Applied Changes
If you need to undo an applied experiment:
# Revert all changes (if not committed)
git reset --hard
git stash apply refs/exps/apply/stash
# Or if already committed
git revert HEAD
DVC creates a backup stash at refs/exps/apply/stash before each apply operation, allowing you to recover your previous state.
Troubleshooting
Merge Conflicts
If applying an experiment causes conflicts:
# DVC will show which files have conflicts
Conflict in params.yaml
# Edit conflicting files
vim params.yaml
# Mark as resolved
git add params.yaml
# Continue working
Missing Experiment
If the experiment doesn’t exist:
dvc exp apply nonexistent-exp
ERROR: experiment 'nonexistent-exp' not found
dvc exp show - View available experimentsdvc exp diff - Compare experiments before applyingdvc exp branch - Promote experiment to a Git branchdvc exp run - Create new experimentsdvc checkout - Checkout DVC-tracked files after applying
