From d16eac687ea8e0a4efa0f478f1a5c937461b5876 Mon Sep 17 00:00:00 2001 From: LiSy Date: Thu, 19 Jun 2025 09:54:37 +0200 Subject: [PATCH 1/4] Add documentation for MLOps Workflow features and best practices --- .../project-dataset/data-versiones-mlops.md | 100 ++++++++++++++++++ 1 file changed, 100 insertions(+) create mode 100644 data-organization/project-dataset/data-versiones-mlops.md diff --git a/data-organization/project-dataset/data-versiones-mlops.md b/data-organization/project-dataset/data-versiones-mlops.md new file mode 100644 index 00000000..0069f8d7 --- /dev/null +++ b/data-organization/project-dataset/data-versiones-mlops.md @@ -0,0 +1,100 @@ +--- +description: >- + The main features, capabilities and usage recommendations of MLOps Workflow + are described in this documentation. +--- + +# 🚀 What is MLOps Workflow in Supervisely + +**MLOps Workflow** in Supervisely helps you: + +1. **Track data changes** – create versions of a project and monitor what changed and when. +2. **Ensure experiment reproducibility** – return to an exact state of data and models. +3. **Collaborate as a team** – see how the project evolved and who ran what. +4. **Understand the process easily** – visualize the entire data/model pipeline. + +## 1. Data Version Control + +- You can save any project state (after uploading, annotating, filtering, etc.) as a **version**. +- Versions are created **automatically** when you run apps like TrainYOLO or other training-related apps. +- This ensures you can **reproduce an experiment** later using the same data state. +- You can also **manually create a version**, but **only if there were changes** in the data. Otherwise, it won’t make sense and the system will prevent it. +- The **only exception** is the first version — you can always create it because there’s no prior state to compare to. +- While versions **can be deleted**, it is **not recommended** — the project history might be lost. +- You may notice **intermediate versions** in the MLOps Workflow graph, but they are just **visual indicators** and do not appear in the project’s version tab. + +## 2. Visualizing the Full Process + +- The interface shows a **graph** – with blocks representing data and applications. +- A block may represent data (projects, datasets, files) **before or after processing**, or an operation/app. +- Example: upload images → save version → train YOLO → get model and report → create a new version. +- Each step is clickable — you can view, download, or recover results at any point. + +## 3. Model Evaluation (Benchmarking) + +- After training, you automatically get a report with: + - mAP, Precision, Recall, IoU + - Inference speed, and more +- These metrics help you compare different models or data setups. +- Easy side-by-side comparison of model versions. + +## 4. Workflow Building – Best Practices + +- Prefer working with **projects**, not just datasets — the graph will be easier to follow. +- **Group related steps** (e.g., apps in folders). +- Add **descriptions** to blocks — teammates will understand your logic faster. +- Avoid **overwriting existing versions** — keep the process traceable. +- When launching apps, **preserve session context**, don’t run everything in one long chain. + +## 5. Getting Started + +You can open the MLOps Workflow interface from: +- The **project menu** +- **Tasks** section +- Directly from the **workspace** + +## 6. Integrating Workflow into Custom Applications + +If you're building your own app using the Python SDK: + +- Initialize with: + ```python + api = Api.from_env() + +This will enable api.app.workflow in your app context. + +Then use methods like: +``` + api.app.workflow.add_input_project(project, version_id=...) +api.app.workflow.add_output_project(...) +``` + +These calls add blocks to the workflow graph showing your app’s input/output. + +There are also methods for datasets, files, sessions, labeling jobs, etc. + +It’s a simple way to visualize your app’s logic in the MLOps Workflow. + +## 7. Why Data Versioning Matters + +To summarize: + +Automatic versions are created by training-related apps (e.g., TrainYOLO). + +- Manual versions require data changes to be saved. + +- Version 1 can always be created — it’s your baseline. + +- Deleting versions is possible but not encouraged. + +- Intermediate versions may appear in the graph view but not in the project’s version list — they are for orientation only. + +## Why MLOps Workflow Is Important + +- Enables you to revert to any data state when needed. + +- Helps analyze performance — what change made the model better? + +- Simplifies teamwork — every step is visible. + +- Speeds up development and delivery — everything is traceable and repeatable. \ No newline at end of file From b9b87908f85696af0a68af0d8c19f3190078d97d Mon Sep 17 00:00:00 2001 From: LiSy Date: Thu, 19 Jun 2025 09:59:31 +0200 Subject: [PATCH 2/4] Add link to "Data Versiones" in the SUMMARY.md --- SUMMARY.md | 1 + 1 file changed, 1 insertion(+) diff --git a/SUMMARY.md b/SUMMARY.md index 7bffe5cc..384d1103 100644 --- a/SUMMARY.md +++ b/SUMMARY.md @@ -77,6 +77,7 @@ * [Collections](data-organization/project-dataset/collections.md) * [Quality Assurance & Statistics](data-organization/project-dataset/quality-assurance-and-statistics/README.md) * [Practical applications of statistics](data-organization/project-dataset/quality-assurance-and-statistics/practical-applications-of-statistics.md) + * [Data Versiones](data-organization/project-dataset/data-versiones-mlops.md) * [MLOps Workflow](data-organization/mlops-workflow.md) * [Team Files](data-organization/team-files/README.md) * [Disk usage & Cleanup](data-organization/storage/README.md) From d5d836f8d396987ee232dd33da4f21b5d7aa8f8b Mon Sep 17 00:00:00 2001 From: LiSy Date: Thu, 10 Jul 2025 13:39:34 +0200 Subject: [PATCH 3/4] Fix page naming --- SUMMARY.md | 2 +- .../{data-versiones-mlops.md => data-versions-mlops.md} | 0 2 files changed, 1 insertion(+), 1 deletion(-) rename data-organization/project-dataset/{data-versiones-mlops.md => data-versions-mlops.md} (100%) diff --git a/SUMMARY.md b/SUMMARY.md index 384d1103..7dec6e79 100644 --- a/SUMMARY.md +++ b/SUMMARY.md @@ -77,7 +77,7 @@ * [Collections](data-organization/project-dataset/collections.md) * [Quality Assurance & Statistics](data-organization/project-dataset/quality-assurance-and-statistics/README.md) * [Practical applications of statistics](data-organization/project-dataset/quality-assurance-and-statistics/practical-applications-of-statistics.md) - * [Data Versiones](data-organization/project-dataset/data-versiones-mlops.md) + * [Data Versions](data-organization/project-dataset/data-versions-mlops.md) * [MLOps Workflow](data-organization/mlops-workflow.md) * [Team Files](data-organization/team-files/README.md) * [Disk usage & Cleanup](data-organization/storage/README.md) diff --git a/data-organization/project-dataset/data-versiones-mlops.md b/data-organization/project-dataset/data-versions-mlops.md similarity index 100% rename from data-organization/project-dataset/data-versiones-mlops.md rename to data-organization/project-dataset/data-versions-mlops.md From 248317c220d42a4ed4b06c49c6e01cc856670ea7 Mon Sep 17 00:00:00 2001 From: LiSy Date: Wed, 16 Jul 2025 15:49:31 +0200 Subject: [PATCH 4/4] Fix formatting in MLOps Workflow documentation --- data-organization/project-dataset/data-versions-mlops.md | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/data-organization/project-dataset/data-versions-mlops.md b/data-organization/project-dataset/data-versions-mlops.md index 0069f8d7..e7d25f73 100644 --- a/data-organization/project-dataset/data-versions-mlops.md +++ b/data-organization/project-dataset/data-versions-mlops.md @@ -1,10 +1,9 @@ --- description: >- - The main features, capabilities and usage recommendations of MLOps Workflow - are described in this documentation. + The main features, capabilities and usage recommendations of MLOps Workflow are described in this documentation. --- -# 🚀 What is MLOps Workflow in Supervisely +# What is MLOps Workflow in Supervisely **MLOps Workflow** in Supervisely helps you: