added tutorial (#185)

* added tutorial * added English version
InnopolisUni · Jul 9, 2023 · b49e712 · b49e712
1 parent ccd0ce8
commit b49e712
Show file tree

Hide file tree

Showing 2 changed files with 191 additions and 0 deletions.
diff --git a/docs/Tutorials - start here/003_Using_Weights_and_Biases_logger_and_Sweeps.md b/docs/Tutorials - start here/003_Using_Weights_and_Biases_logger_and_Sweeps.md
@@ -0,0 +1,94 @@
+# Using the Weights and Biases Logger
+[Video](https://www.youtube.com/watch?v=zgt2YFNEOMY&ab_channel=innofw) about using Weights and Biases in the `innofw` framework.
+
+Add logger to the experiment config:
+
+```yaml
+# @package _global_
+task: "image-segmentation"
+defaults:
+  - override /models: semantic-segmentation/unet_smp.yaml
+  - override /datasets: semantic-segmentation/stroke_dataset.yaml
+  - override /losses: semantic-segmentation/focal_tversky_loss.yaml
+  - override /loggers: wandb  # <-- add this line to use wandb logger
+```
+
+## Sweep
+You can get acquainted with what a wandb sweep is in the [documentation](https://docs.wandb.ai/guides/sweeps).
+
+Here we will consider how and where to create configuration files and how to run a sweep in the framework.
+
+1. You need to create a configuration file. Example of a configuration file for a sweep (let's say the file: config/experiments/sweeps/SO_120323_example.yaml):
+
+
+```yaml
+# wandb sweep config
+program: train.py
+method: bayes
+metric:
+  name: val_MulticlassJaccardIndex
+  goal: maximize
+
+name: param_tuning_deeplabv3plus
+project: uw_madison
+
+parameters:
+
+  experiments:
+    value: "KG_260423_inwrg21_deeplab_uwmadison.yaml"
+
+  augmentations_train:
+    values: [none, 'position/random_horizontal_flip.yaml']
+
+  batch_size:
+    values: [ 64, 128, 256 ]
+
+  models.encoder_name:
+    values: ['resnet34', 'resnet50', 'resnet101']
+
+  optimizers:
+    values: ['lion.yaml', 'adam.yaml']
+
+  optimizers.lr:
+    min: 0.0035
+    max: 0.0070
+
+  epochs:
+    values: [15, 20]
+
+command:
+  - ${env}
+  - python
+  - ${program}
+  - ${args_no_hyphens}
+```
+
+After this, you need to create a job using the `wandb sweep` command, providing the path to the file. Example of the command:
+
+```bash
+wandb sweep config/experiments/sweeps/config/experiments/sweeps/SO_120323_example.yaml
+```
+Expected command output:
+
+```bash
+wandb: Creating sweep from: config/experiments/sweeps/KG_280423_obwg91_sweep_test.yaml
+wandb: Created sweep with ID: tx1o222v
+wandb: View sweep at: https://wandb.ai/qb-1/uw_madison/sweeps/tx1o222v
+wandb: Run sweep agent with: wandb agent qb-1/uw_madison/tx1o222v
+```
+
+You need to copy this part: `wandb agent qb-1/uw_madison/tx1o222v`
+
+3. Now you can start the processes to carry out tasks from the job. Here is an example of starting two processes of our task on gpu=0, 2:
+
+```
+CUDA_VISIBLE_DEVICES=0 wandb agent qazybi/uwmadison/329fj23
+```
+
+```bash
+CUDA_VISIBLE_DEVICES=2 wandb agent qazybi/uwmadison/329fj23
+```
+
+In case you encounter a `PermissionError`, try running the process in sudo mode:
+```bash
+CUDA_VISIBLE_DEVICES=2 sudo -E env "PATH=$PATH" wandb agent qazybi/uwmadison/329fj23```
diff --git a/docs/Tutorials - start here/ru/003_Using_Weights_and_Biases_logger_and_Sweeps.md b/docs/Tutorials - start here/ru/003_Using_Weights_and_Biases_logger_and_Sweeps.md
@@ -0,0 +1,97 @@
+# Использование логгера Weights and Biases
+
+[Видео](https://www.youtube.com/watch?v=zgt2YFNEOMY&ab_channel=innofw) про использование Weights and Biases во фреймворке `innofw`.
+
+добавить логгер в конфиг эксперимента
+
+```yaml
+# @package _global_
+task: "image-segmentation"
+defaults:
+  - override /models: semantic-segmentation/unet_smp.yaml
+  - override /datasets: semantic-segmentation/stroke_dataset.yaml
+  - override /losses: semantic-segmentation/focal_tversky_loss.yaml
+  - override /loggers: wandb  # <-- add this line to use wandb logger
+```
+
+## Sweep
+
+с тем что такое wandb sweep можно ознакомиться по [документации](https://docs.wandb.ai/guides/sweeps)
+
+Здесь же мы рассмотрим как и где создать конфигурационные файлы и как запустить sweep во фреймворке.
+
+1. Нужно создать конфигурационный файл. Пример конфигурационного файла для sweep(допустим файл: `config/experiments/sweeps/SO_120323_example.yaml`):
+
+    ```yaml
+    # wandb sweep config
+    program: train.py
+    method: bayes
+    metric:
+      name: val_MulticlassJaccardIndex
+      goal: maximize
+    
+    name: param_tuning_deeplabv3plus
+    project: uw_madison
+    
+    parameters:
+    
+      experiments:
+        value: "KG_260423_inwrg21_deeplab_uwmadison.yaml"
+    
+      augmentations_train:
+        values: [none, 'position/random_horizontal_flip.yaml']
+    
+      batch_size:
+        values: [ 64, 128, 256 ]
+    
+      models.encoder_name:
+        values: ['resnet34', 'resnet50', 'resnet101']
+    
+      optimizers:
+        values: ['lion.yaml', 'adam.yaml']
+    
+      optimizers.lr:
+        min: 0.0035
+        max: 0.0070
+    
+      epochs:
+        values: [15, 20]
+    
+    command:
+      - ${env}
+      - python
+      - ${program}
+      - ${args_no_hyphens}
+    ```
+
+2. После этого надо будет создать job командой `wandb sweep` передав путь до файла. Пример команды:
+
+    ```bash
+    wandb sweep config/experiments/sweeps/config/experiments/sweeps/SO_120323_example.yaml
+    ```
+
+    Примерный вывод команды:
+
+    ```bash
+    wandb: Creating sweep from: config/experiments/sweeps/KG_280423_obwg91_sweep_test.yaml
+    wandb: Created sweep with ID: tx1o222v
+    wandb: View sweep at: https://wandb.ai/qb-1/uw_madison/sweeps/tx1o222v
+    wandb: Run sweep agent with: wandb agent qb-1/uw_madison/tx1o222v
+    ```
+
+    Нужно скопировать часть: `wandb agent qb-1/uw_madison/tx1o222v`
+
+3. Теперь можно запустить процессы, чтобы выполнялись задачи с job-а. Вот пример запуска двух процессов нашей задачи на гпу=0, 2: 
+
+    ```bash
+    CUDA_VISIBLE_DEVICES=0 wandb agent qazybi/uwmadison/329fj23
+    ```
+
+    ```bash
+    CUDA_VISIBLE_DEVICES=2 wandb agent qazybi/uwmadison/329fj23
+    ```
+
+    В случае если возникает ошибка `PermissionError`, попробуйте запустить процесс в режиме sudo:
+    ```bash
+    CUDA_VISIBLE_DEVICES=2 sudo -E env "PATH=$PATH" wandb agent qazybi/uwmadison/329fj23
+    ```