exp show

Displays your experiments in a customizable table or parallel coordinates plot.

Press q to exit.

Synopsis

usage: dvc exp show [-h] [-q | -v] [-a] [-T] [-A] [-n <num>]
                    [--no-pager] [--drop <regex>]
                    [--keep <regex>] [--param-deps]
                    [--sort-by <metric/param>]
                    [--sort-order {asc,desc}] [--sha]
                    [--json] [--csv] [--md] [--precision <n>]
                    [--pcp] [--only-changed]

Description

Displays experiments and checkpoints in a detailed table which includes their parent and name (or hash), as well as project metrics and parameters. Only the experiments derived from the Git HEAD are shown by default but all experiments can be included with the --all-commits option. Example:

$ dvc exp show

 ───────────────────────────────────────────────────────────────────
  neutral:**Experiment**      metric:**avg_prec**   metric:**roc_auc**   param:**train.n_est**  param:**train.min_split**
 ───────────────────────────────────────────────────────────────────
  workspace        0.56191   0.93345   50           2
  master           0.55259   0.91536   50           2
  ├── exp-bfe64    0.57833   0.95555   50           8
  └── exp-ad5b1    0.56191   0.93345   50           2
 ───────────────────────────────────────────────────────────────────

Your terminal will enter a paginated screen by default, which you can typically exit by typing Q. Use --no-pager to print the table to standard output.

By default, the printed experiments table will include columns for all metrics and params from the entire project. The --only-changed, --drop, --keep, and other options can determine which ones should be displayed.

Experiments in the table are first grouped (by parent commit). They are then sorted inside each group, chronologically by default. The --sort-by and --sort-order options can change this ordering, based on any single, visible metric or param.

When the --pcp option is passed, an interactive parallel coordinates plot will be generated using the same data from the table.

Parallel Coordinates Plot

Options

-a, --all-branches - include experiments derived from all Git branches, as well as from the last commit (HEAD). Note that this can be combined with -T below, for example using the -aT flags.
-T, --all-tags - include experiments derived from all Git tags, as well as from the last commit. Note that this can be combined with -a above, for example using the -aT flags.
-A, --all-commits - include experiments derived from all Git commits, as well as from the last one. This prints all experiments in the project.
-n <num>, --num <num> - show the last num commits from HEAD.
--no-pager - do not enter the pager screen. Writes the entire table to standard output. Useful to redirect the output to a file, or use your own paginator.
--param-deps - include only parameters that are stage dependencies.
--only-changed - show only parameters and metrics with values that vary across experiments.
--drop <regex> - remove the matching columns. This option has higher priority than --only-changed. If both options are combined, --drop will remove matching columns even if their values vary across experiments.
--keep <regex> - prevent the matching columns to be removed by any of the other options, including --only-changed and --drop.
--sort-by <name> - sort experiments by the specified metric or param (name). Only one visible column (either metric or param) can be used for sorting. This only affects the ordering of experiments derived from the same parent commit. Parent commits are always sorted chronologically.
--sort-order {asc,desc} - sort order to use with --sort-by. Defaults to ascending (asc).
--sha - display Git commit (SHA) hashes instead of branch, tag, or experiment names.
--json - prints the command's output in easily parsable JSON format, instead of a human-readable table.
--csv - prints the command's output in CSV format instead of a human-readable table.
--md - prints the command's output in Markdown table format.
--precision <n> - round decimal values to n digits of precision (5 by default). Applies to metrics only.
-h, --help - prints the usage/help message, and exit.
-q, --quiet - do not write anything to standard output. Exit with 0 if no problems arise, otherwise 1.
-v, --verbose - displays detailed tracing information.
--pcp - generates an interactive parallel coordinates plot from the table.
-o <folder>, --out <folder> - when used with --pcp, specifies a destination folder of the plot. By default its dvc_plots.
--open - when used with --pcp, opens the generated plot in a browser automatically. You can enable dvc config plots.auto_open to make this the default behavior.

Examples

This example is based on our Get Started, where you can find the actual source code.

Let's say we have run 3 experiments in our project. The basic usage shows the workspace (Git working tree) and experiments derived from HEAD (master branch in this case), and all of their metrics and params (scroll right to see all):

$ dvc exp show

 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**avg_prec**   metric:**roc_auc**   param:**prepare.split**   param:**prepare.seed**   param:**featurize.max_features**   param:**featurize.ngrams**   param:**train.seed**   param:**train.n_est**   param:**train.min_split**
 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.60405    0.9608   0.2             20170428       3000                     2                  20170428     100           64
  master                    May 29, 2021    0.60405    0.9608   0.2             20170428       3000                     2                  20170428     100           64
  ├── d384680 [exp-bc055]   08:03 PM        0.51799   0.92333   0.2             20170428       500                      2                  20170428     100           64
  ├── 6b338f8 [exp-3315b]   08:03 PM        0.58589     0.945   0.2             20170428       2000                     2                  20170428     100           64
  └── d7fdde2 [exp-1b262]   08:03 PM        0.56447   0.94713   0.2             20170428       1500                     2                  20170428     100           64
 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

You can exit this screen with Q, typically.

As a quick way of reducing noise, --only-changed will drop any column with values that do not change across experiments:

$ dvc exp show --only-changed

 ──────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**avg_prec**   metric:**roc_auc**   param:**featurize.max_features**
 ──────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.60405    0.9608   3000
  master                    May 29, 2021    0.60405    0.9608   3000
  ├── d7fdde2 [exp-1b262]   08:03 PM        0.56447   0.94713   1500
  ├── 6b338f8 [exp-3315b]   08:03 PM        0.58589     0.945   2000
  └── d384680 [exp-bc055]   08:03 PM        0.51799   0.92333   500
 ──────────────────────────────────────────────────────────────────────────────────────

You can also use --drop to filter specific columns:

$ dvc exp show --drop prepare

 ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**avg_prec**   metric:**roc_auc**   param:**featurize.max_features**   param:**featurize.ngrams**   param:**train.seed**   param:**train.n_est**   param:**train.min_split**
 ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.60405    0.9608   3000                     2                  20170428     100           64
  master                    May 29, 2021    0.60405    0.9608   3000                     2                  20170428     100           64
  ├── 6b338f8 [exp-3315b]   08:03 PM        0.58589     0.945   2000                     2                  20170428     100           64
  ├── d384680 [exp-bc055]   08:03 PM        0.51799   0.92333   500                      2                  20170428     100           64
  └── d7fdde2 [exp-1b262]   08:03 PM        0.56447   0.94713   1500                     2                  20170428     100           64
 ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

You can use regex to match columns. For example, to remove multiple columns:

$ dvc exp show --drop 'avg_prec|train.min_split'

 ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**roc_auc**   param:**prepare.split**   param:**prepare.seed**   param:**featurize.max_features**   param:**featurize.ngrams**   param:**train.seed**   param:**train.n_est**
 ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.9608   0.2             20170428       3000                     2                  20170428     100
  master                    May 29, 2021    0.9608   0.2             20170428       3000                     2                  20170428     100
  ├── d384680 [exp-bc055]   Dec 17, 2021   0.92333   0.2             20170428       500                      2                  20170428     100
  ├── d7fdde2 [exp-1b262]   Dec 17, 2021   0.94713   0.2             20170428       1500                     2                  20170428     100
  └── 6b338f8 [exp-3315b]   Dec 17, 2021     0.945   0.2             20170428       2000                     2                  20170428     100
 ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

If combined --only-changed has the least priority, --drop comes next, and --keep has the last word:

$ dvc exp show --only-changed --drop Created --keep 'train.(?!seed)'

 ───────────────────────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                metric:**avg_prec**   metric:**roc_auc**   param:**featurize.max_features**   param:**train.n_est**   param:**train.min_split**
 ───────────────────────────────────────────────────────────────────────────────────────────────────────
  workspace                  0.60405    0.9608   3000                     100           64
  master                     0.60405    0.9608   3000                     100           64
  ├── d384680 [exp-bc055]    0.51799   0.92333   500                      100           64
  ├── 6b338f8 [exp-3315b]    0.58589     0.945   2000                     100           64
  └── d7fdde2 [exp-1b262]    0.56447   0.94713   1500                     100           64
 ───────────────────────────────────────────────────────────────────────────────────────────────────────

Sort experiments by the roc_auc metric, in descending order:

$ dvc exp show --only-changed --sort-by=roc_auc --sort-order desc

 ──────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**avg_prec**   metric:**roc_auc**   param:**featurize.max_features**
 ──────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.60405    0.9608   3000
  master                    May 29, 2021    0.60405    0.9608   3000
  ├── d7fdde2 [exp-1b262]   08:03 PM        0.56447   0.94713   1500
  ├── 6b338f8 [exp-3315b]   08:03 PM        0.58589     0.945   2000
  └── d384680 [exp-bc055]   08:03 PM        0.51799   0.92333   500
 ──────────────────────────────────────────────────────────────────────────────────────

To see all experiments throughout the Git history:

$ dvc exp show --all-commits --only-changed --sort-by=roc_auc

 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  neutral:**Experiment**                neutral:**Created**        metric:**avg_prec**   metric:**roc_auc**   param:**featurize.max_features**   param:**featurize.ngrams**   param:**train.n_est**   param:**train.min_split**
 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  workspace                 -               0.60405    0.9608   3000                     2                  100           64
  try-large-dataset         Jun 01, 2021    0.67038   0.96693   3000                     2                  100           64
  master                    May 29, 2021    0.60405    0.9608   3000                     2                  100           64
  ├── d384680 [exp-bc055]   08:03 PM        0.51799   0.92333   500                      2                  100           64
  ├── 6b338f8 [exp-3315b]   08:03 PM        0.58589     0.945   2000                     2                  100           64
  └── d7fdde2 [exp-1b262]   08:03 PM        0.56447   0.94713   1500                     2                  100           64
  cc51022                   May 28, 2021    0.55259   0.91536   1500                     2                  50            2
  7ab3585                   May 27, 2021    0.52048    0.9032   1500                     2                  50            2
  53b2d9d                   May 25, 2021    0.52048    0.9032   500                      1                  50            2
  872cd6c                   May 24, 2021          -         -   500                      1                  50            2
  8188b34                   May 23, 2021          -         -   500                      1                  50            2
  9244ec3                   May 22, 2021          -         -   500                      1                  50            2
  08a3b89                   May 21, 2021          -         -   -                        -                  -             -
  16ba2cd                   May 20, 2021          -         -   -                        -                  -             -
  f0c0269                   May 18, 2021          -         -   -                        -                  -             -
  3e07290                   May 17, 2021          -         -   -                        -                  -             -
  90b2aea                   May 16, 2021          -         -   -                        -                  -             -
 ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

Note that in this example, Git commits remain in chronological order. The sorting only applies to experiment groups (sharing a parent commit).

Example: Parallel coordinates plot (PCP)

To generate an interactive parallel coordiantes plot based on the experiments and their parameters:

$ dvc exp show --all-branches --pcp

ref pcp default Parallel Coordinates Plot

Using --sort-by will reorder the plot experiments as expected, and determine the color of the lines that represent them:

$ dvc exp show --all-branches --pcp --sort-by roc_auc

ref pcp sortby Colorized by roc_auc

Combine with other flags for further filtering:

$ dvc exp show --all-branches --pcp --sort-by roc_auc
               --exclude-metrics avg_prec

ref pcp filter Excluded avg_prec column

📖 See Metrics, Parameters, and Plots for an introduction to parameters, metrics, plots.