# [](https://colab.research.google.com/drive/1a0pSD-1tWhMmeJeeoyZM1A-HCW3yf1xR?usp=sharing) [](https://agarwl.github.io/rliable) [](https://ai.googleblog.com/2021/11/rliable-towards-reliable-evaluation.html)
`rliable` is an open-source Python library for reliable evaluation, even with a *handful
of runs*, on reinforcement learning and machine learnings benchmarks.
| **Desideratum** | **Current evaluation approach** | **Our Recommendation** |
| --------------------------------- | ----------- | --------- |
| Uncertainty in aggregate performance | **Point estimates**: <ul> <li> Ignore statistical uncertainty </li> <li> Hinder *results reproducibility* </li></ul> | Interval estimates using **stratified bootstrap confidence intervals** (CIs) |
|Performance variability across tasks and runs| **Tables with task mean scores**: <ul><li> Overwhelming beyond a few tasks </li> <li> Standard deviations frequently omitted </li> <li> Incomplete picture for multimodal and heavy-tailed distributions </li> </ul> | **Score distributions** (*performance profiles*): <ul> <li> Show tail distribution of scores on combined runs across tasks </li> <li> Allow qualitative comparisons </li> <li> Easily read any score percentile </li> </ul>|
|Aggregate metrics for summarizing benchmark performance | **Mean**: <ul><li> Often dominated by performance on outlier tasks </li></ul> **Median**: <ul> <li> Statistically inefficient (requires a large number of runs to claim improvements) </li> <li> Poor indicator of overall performance: 0 scores on nearly half the tasks doesn't change it </li> </ul>| **Interquartile Mean (IQM)** across all runs: <ul> <li> Performance on middle 50% of combined runs </li> <li> Robust to outlier scores but more statistically efficient than median </li> </ul> To show other aspects of performance gains, report *Probability of improvement* and *Optimality gap* |
`rliable` provides support for:
* Stratified Bootstrap Confidence Intervals (CIs)
* Performance Profiles (with plotting functions)
* Aggregate metrics
* Interquartile Mean (IQM) across all runs
* Optimality Gap
* Probability of Improvement
<div align="left">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/aggregate_metric.png">
</div>
## Interactive colab
We provide a colab at [bit.ly/statistical_precipice_colab](https://colab.research.google.com/drive/1a0pSD-1tWhMmeJeeoyZM1A-HCW3yf1xR?usp=sharing),
which shows how to use the library with examples of published algorithms on
widely used benchmarks including Atari 100k, ALE, DM Control and Procgen.
### Data for individual runs on Atari 100k, ALE, DM Control and Procgen
You can access the data for individual runs using the public GCP bucket here (you might need to sign in with your
gmail account to use Gcloud) : https://console.cloud.google.com/storage/browser/rl-benchmark-data.
The interactive colab above also allows you to access the data programatically.
### Paper
For more details, refer to the accompanying **NeurIPS 2021** paper (**Outstanding Paper** Award):
[Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://arxiv.org/pdf/2108.13264.pdf).
### Installation
To install `rliable`, run:
```python
pip install -U rliable
```
To install latest version of `rliable` as a package, run:
```python
pip install git+https://github.com/google-research/rliable
```
To import `rliable`, we suggest:
```python
from rliable import library as rly
from rliable import metrics
from rliable import plot_utils
```
### Aggregate metrics with 95% Stratified Bootstrap CIs
##### IQM, Optimality Gap, Median, Mean
```python
algorithms = ['DQN (Nature)', 'DQN (Adam)', 'C51', 'REM', 'Rainbow',
'IQN', 'M-IQN', 'DreamerV2']
# Load ALE scores as a dictionary mapping algorithms to their human normalized
# score matrices, each of which is of size `(num_runs x num_games)`.
atari_200m_normalized_score_dict = ...
aggregate_func = lambda x: np.array([
metrics.aggregate_median(x),
metrics.aggregate_iqm(x),
metrics.aggregate_mean(x),
metrics.aggregate_optimality_gap(x)])
aggregate_scores, aggregate_score_cis = rly.get_interval_estimates(
atari_200m_normalized_score_dict, aggregate_func, reps=50000)
fig, axes = plot_utils.plot_interval_estimates(
aggregate_scores, aggregate_score_cis,
metric_names=['Median', 'IQM', 'Mean', 'Optimality Gap'],
algorithms=algorithms, xlabel='Human Normalized Score')
```
<div align="left">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/ale_interval_estimates.png">
</div>
##### Probability of Improvement
```python
# Load ProcGen scores as a dictionary containing pairs of normalized score
# matrices for pairs of algorithms we want to compare
procgen_algorithm_pairs = {.. , 'x,y': (score_x, score_y), ..}
average_probabilities, average_prob_cis = rly.get_interval_estimates(
procgen_algorithm_pairs, metrics.probability_of_improvement, reps=2000)
plot_utils.plot_probability_of_improvement(average_probabilities, average_prob_cis)
```
<div align="center">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/procgen_probability_of_improvement.png">
</div>
#### Sample Efficiency Curve
```python
algorithms = ['DQN (Nature)', 'DQN (Adam)', 'C51', 'REM', 'Rainbow',
'IQN', 'M-IQN', 'DreamerV2']
# Load ALE scores as a dictionary mapping algorithms to their human normalized
# score matrices across all 200 million frames, each of which is of size
# `(num_runs x num_games x 200)` where scores are recorded every million frame.
ale_all_frames_scores_dict = ...
frames = np.array([1, 10, 25, 50, 75, 100, 125, 150, 175, 200]) - 1
ale_frames_scores_dict = {algorithm: score[:, :, frames] for algorithm, score
in ale_all_frames_scores_dict.items()}
iqm = lambda scores: np.array([metrics.aggregate_iqm(scores[..., frame])
for frame in range(scores.shape[-1])])
iqm_scores, iqm_cis = rly.get_interval_estimates(
ale_frames_scores_dict, iqm, reps=50000)
plot_utils.plot_sample_efficiency_curve(
frames+1, iqm_scores, iqm_cis, algorithms=algorithms,
xlabel=r'Number of Frames (in millions)',
ylabel='IQM Human Normalized Score')
```
<div align="center">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/ale_legend.png">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/atari_sample_efficiency_iqm.png">
</div>
### Performance Profiles
```python
# Load ALE scores as a dictionary mapping algorithms to their human normalized
# score matrices, each of which is of size `(num_runs x num_games)`.
atari_200m_normalized_score_dict = ...
# Human normalized score thresholds
atari_200m_thresholds = np.linspace(0.0, 8.0, 81)
score_distributions, score_distributions_cis = rly.create_performance_profile(
atari_200m_normalized_score_dict, atari_200m_thresholds)
# Plot score distributions
fig, ax = plt.subplots(ncols=1, figsize=(7, 5))
plot_utils.plot_performance_profiles(
score_distributions, atari_200m_thresholds,
performance_profile_cis=score_distributions_cis,
colors=dict(zip(algorithms, sns.color_palette('colorblind'))),
xlabel=r'Human Normalized Score $(\tau)$',
ax=ax)
```
<div align="center">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/ale_legend.png">
<img src="https://raw.githubusercontent.com/google-research/rliable/master/images/ale_score_distributions_new.png">
</div>
The above profile can also be plotted with non-linear scaling as follows:
```python
plot_utils.plot_performance_profiles(
perf_prof_atari_200m, atari_200m_tau,
performance_profile_cis=perf_prof_atari_200m_cis,
use_non_linear_scaling=True,
xticks = [0.0, 0.5, 1.0, 2.0, 4.0, 8.0]
colors=dict(zip(algorithms, sns.color_palette('colorblind'))),
xlabel=r'Human Normalized Score $(\tau)$',
ax=ax)
```
### Dependencies
The code was tested under `Python>=3.7` and uses these packages:
- arch == 5.3.0
- scipy >= 1.7.0
- numpy >= 0.9.0
- absl-py >= 1.16.4
- seaborn >= 0.11.2
Citing
------
If you find this open source release useful, please reference in your paper:
@article{agarwal2021deep,
title={Deep Reinforcement Learning at the Edge of the Statistical Precipice},
author={Agarwal, Rishabh and Schwarzer, Max and Castro, Pablo Samuel
and Courville, Aaron and Bellemare, Marc G},
journal={Advances in Neural Information Processing Systems},
year={2021}
}
Disclaimer: This is not an official Google product.
Raw data
{
"_id": null,
"home_page": "https://github.com/google-research/rliable",
"name": "rliable",
"maintainer": null,
"docs_url": null,
"requires_python": null,
"maintainer_email": null,
"keywords": "benchmarking, evaluation, reproducibility, research, reinforcement, machine, learning, research",
"author": "Rishabh Agarwal",
"author_email": null,
"download_url": "https://files.pythonhosted.org/packages/d2/ca/b9e0e274d4e53f39654ac02eac43cd707fd12273f69f12b220c00baba020/rliable-1.2.0.tar.gz",
"platform": null,
"description": "\n# [](https://colab.research.google.com/drive/1a0pSD-1tWhMmeJeeoyZM1A-HCW3yf1xR?usp=sharing) [](https://agarwl.github.io/rliable) [](https://ai.googleblog.com/2021/11/rliable-towards-reliable-evaluation.html)\n\n`rliable` is an open-source Python library for reliable evaluation, even with a *handful\nof runs*, on reinforcement learning and machine learnings benchmarks. \n| **Desideratum** | **Current evaluation approach** | **Our Recommendation** |\n| --------------------------------- | ----------- | --------- |\n| Uncertainty in aggregate performance | **Point estimates**: <ul> <li> Ignore statistical uncertainty </li> <li> Hinder *results reproducibility* </li></ul> | Interval estimates using **stratified bootstrap confidence intervals** (CIs) |\n|Performance variability across tasks and runs| **Tables with task mean scores**: <ul><li> Overwhelming beyond a few tasks </li> <li> Standard deviations frequently omitted </li> <li> Incomplete picture for multimodal and heavy-tailed distributions </li> </ul> | **Score distributions** (*performance profiles*): <ul> <li> Show tail distribution of scores on combined runs across tasks </li> <li> Allow qualitative comparisons </li> <li> Easily read any score percentile </li> </ul>|\n|Aggregate metrics for summarizing benchmark performance | **Mean**: <ul><li> Often dominated by performance on outlier tasks </li></ul> **Median**: <ul> <li> Statistically inefficient (requires a large number of runs to claim improvements) </li> <li> Poor indicator of overall performance: 0 scores on nearly half the tasks doesn't change it </li> </ul>| **Interquartile Mean (IQM)** across all runs: <ul> <li> Performance on middle 50% of combined runs </li> <li> Robust to outlier scores but more statistically efficient than median </li> </ul> To show other aspects of performance gains, report *Probability of improvement* and *Optimality gap* |\n\n`rliable` provides support for:\n\n * Stratified Bootstrap Confidence Intervals (CIs)\n * Performance Profiles (with plotting functions)\n * Aggregate metrics\n * Interquartile Mean (IQM) across all runs\n * Optimality Gap\n * Probability of Improvement\n\n<div align=\"left\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/aggregate_metric.png\">\n</div>\n\n## Interactive colab\nWe provide a colab at [bit.ly/statistical_precipice_colab](https://colab.research.google.com/drive/1a0pSD-1tWhMmeJeeoyZM1A-HCW3yf1xR?usp=sharing),\nwhich shows how to use the library with examples of published algorithms on\nwidely used benchmarks including Atari 100k, ALE, DM Control and Procgen.\n\n### Data for individual runs on Atari 100k, ALE, DM Control and Procgen\n\nYou can access the data for individual runs using the public GCP bucket here (you might need to sign in with your\ngmail account to use Gcloud) : https://console.cloud.google.com/storage/browser/rl-benchmark-data.\nThe interactive colab above also allows you to access the data programatically.\n\n### Paper\nFor more details, refer to the accompanying **NeurIPS 2021** paper (**Outstanding Paper** Award):\n[Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://arxiv.org/pdf/2108.13264.pdf).\n\n\n### Installation\n\nTo install `rliable`, run:\n```python\npip install -U rliable\n```\n\nTo install latest version of `rliable` as a package, run:\n\n```python\npip install git+https://github.com/google-research/rliable\n```\n\nTo import `rliable`, we suggest:\n\n```python\nfrom rliable import library as rly\nfrom rliable import metrics\nfrom rliable import plot_utils\n```\n\n### Aggregate metrics with 95% Stratified Bootstrap CIs\n\n\n##### IQM, Optimality Gap, Median, Mean\n```python\nalgorithms = ['DQN (Nature)', 'DQN (Adam)', 'C51', 'REM', 'Rainbow',\n 'IQN', 'M-IQN', 'DreamerV2']\n# Load ALE scores as a dictionary mapping algorithms to their human normalized\n# score matrices, each of which is of size `(num_runs x num_games)`.\natari_200m_normalized_score_dict = ...\naggregate_func = lambda x: np.array([\n metrics.aggregate_median(x),\n metrics.aggregate_iqm(x),\n metrics.aggregate_mean(x),\n metrics.aggregate_optimality_gap(x)])\naggregate_scores, aggregate_score_cis = rly.get_interval_estimates(\n atari_200m_normalized_score_dict, aggregate_func, reps=50000)\nfig, axes = plot_utils.plot_interval_estimates(\n aggregate_scores, aggregate_score_cis,\n metric_names=['Median', 'IQM', 'Mean', 'Optimality Gap'],\n algorithms=algorithms, xlabel='Human Normalized Score')\n```\n\n<div align=\"left\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/ale_interval_estimates.png\">\n</div>\n\n##### Probability of Improvement\n```python\n# Load ProcGen scores as a dictionary containing pairs of normalized score\n# matrices for pairs of algorithms we want to compare\nprocgen_algorithm_pairs = {.. , 'x,y': (score_x, score_y), ..}\naverage_probabilities, average_prob_cis = rly.get_interval_estimates(\n procgen_algorithm_pairs, metrics.probability_of_improvement, reps=2000)\nplot_utils.plot_probability_of_improvement(average_probabilities, average_prob_cis)\n```\n<div align=\"center\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/procgen_probability_of_improvement.png\">\n</div>\n\n#### Sample Efficiency Curve\n```python\nalgorithms = ['DQN (Nature)', 'DQN (Adam)', 'C51', 'REM', 'Rainbow',\n 'IQN', 'M-IQN', 'DreamerV2']\n# Load ALE scores as a dictionary mapping algorithms to their human normalized\n# score matrices across all 200 million frames, each of which is of size\n# `(num_runs x num_games x 200)` where scores are recorded every million frame.\nale_all_frames_scores_dict = ...\nframes = np.array([1, 10, 25, 50, 75, 100, 125, 150, 175, 200]) - 1\nale_frames_scores_dict = {algorithm: score[:, :, frames] for algorithm, score\n in ale_all_frames_scores_dict.items()}\niqm = lambda scores: np.array([metrics.aggregate_iqm(scores[..., frame])\n for frame in range(scores.shape[-1])])\niqm_scores, iqm_cis = rly.get_interval_estimates(\n ale_frames_scores_dict, iqm, reps=50000)\nplot_utils.plot_sample_efficiency_curve(\n frames+1, iqm_scores, iqm_cis, algorithms=algorithms,\n xlabel=r'Number of Frames (in millions)',\n ylabel='IQM Human Normalized Score')\n```\n<div align=\"center\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/ale_legend.png\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/atari_sample_efficiency_iqm.png\">\n</div>\n\n### Performance Profiles\n\n```python\n# Load ALE scores as a dictionary mapping algorithms to their human normalized\n# score matrices, each of which is of size `(num_runs x num_games)`.\natari_200m_normalized_score_dict = ...\n# Human normalized score thresholds\natari_200m_thresholds = np.linspace(0.0, 8.0, 81)\nscore_distributions, score_distributions_cis = rly.create_performance_profile(\n atari_200m_normalized_score_dict, atari_200m_thresholds)\n# Plot score distributions\nfig, ax = plt.subplots(ncols=1, figsize=(7, 5))\nplot_utils.plot_performance_profiles(\n score_distributions, atari_200m_thresholds,\n performance_profile_cis=score_distributions_cis,\n colors=dict(zip(algorithms, sns.color_palette('colorblind'))),\n xlabel=r'Human Normalized Score $(\\tau)$',\n ax=ax)\n```\n<div align=\"center\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/ale_legend.png\">\n <img src=\"https://raw.githubusercontent.com/google-research/rliable/master/images/ale_score_distributions_new.png\">\n</div>\n\nThe above profile can also be plotted with non-linear scaling as follows:\n\n```python\nplot_utils.plot_performance_profiles(\n perf_prof_atari_200m, atari_200m_tau,\n performance_profile_cis=perf_prof_atari_200m_cis,\n use_non_linear_scaling=True,\n xticks = [0.0, 0.5, 1.0, 2.0, 4.0, 8.0]\n colors=dict(zip(algorithms, sns.color_palette('colorblind'))),\n xlabel=r'Human Normalized Score $(\\tau)$',\n ax=ax)\n```\n\n\n### Dependencies\nThe code was tested under `Python>=3.7` and uses these packages:\n\n- arch == 5.3.0\n- scipy >= 1.7.0\n- numpy >= 0.9.0\n- absl-py >= 1.16.4\n- seaborn >= 0.11.2\n\nCiting\n------\nIf you find this open source release useful, please reference in your paper:\n\n @article{agarwal2021deep,\n title={Deep Reinforcement Learning at the Edge of the Statistical Precipice},\n author={Agarwal, Rishabh and Schwarzer, Max and Castro, Pablo Samuel\n and Courville, Aaron and Bellemare, Marc G},\n journal={Advances in Neural Information Processing Systems},\n year={2021}\n }\n\nDisclaimer: This is not an official Google product.\n",
"bugtrack_url": null,
"license": "Apache 2.0",
"summary": "rliable: Reliable evaluation on reinforcement learning and machine learning benchmarks.",
"version": "1.2.0",
"project_urls": {
"Bug Reports": "https://github.com/google-research/rliable/issues",
"Documentation": "https://github.com/google-research/rliable",
"Homepage": "https://github.com/google-research/rliable",
"Source": "https://github.com/google-research/rliable"
},
"split_keywords": [
"benchmarking",
" evaluation",
" reproducibility",
" research",
" reinforcement",
" machine",
" learning",
" research"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "d2cab9e0e274d4e53f39654ac02eac43cd707fd12273f69f12b220c00baba020",
"md5": "3982465ad6dc19427d6e467abf5b7187",
"sha256": "72789d9147d7c56e6efa812f9dffedcef44993a866ec08d75506ac7c1fe69cd5"
},
"downloads": -1,
"filename": "rliable-1.2.0.tar.gz",
"has_sig": false,
"md5_digest": "3982465ad6dc19427d6e467abf5b7187",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 21175,
"upload_time": "2024-08-12T20:50:58",
"upload_time_iso_8601": "2024-08-12T20:50:58.667404Z",
"url": "https://files.pythonhosted.org/packages/d2/ca/b9e0e274d4e53f39654ac02eac43cd707fd12273f69f12b220c00baba020/rliable-1.2.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-08-12 20:50:58",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "google-research",
"github_project": "rliable",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "rliable"
}
JSON
