### Example Configuration Parameter

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Defines a configuration parameter with its type, default value, and a descriptive comment. The comment must start with the parameter's type.

```yaml
frame_limit: 15 # int; tracklets shorter than this number of frames will be discarded
```

--------------------------------

### Run Episode with Parameter Updates

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Start a new training episode with modified parameters. Use `parameters_update` to specify changes to configuration files like 'general' or 'ssl'.

```python
project.run_episode(
    'episode_5',
    parameters_update={
        'general': {'ssl': ['contrastive']},
        'ssl': {'contrastive': {'ssl_weight': 0.01}},
    },
)
```

--------------------------------

### Initialize DLC2Action Project

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Instantiate a Project object to begin your experiments. Ensure the project name is provided, and optionally, the projects path if it was used during project creation.

```python
from dlc2action.project import Project
project = Project('project_name')
```

--------------------------------

### Run Episode Loading Previous Model and Parameters

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Create a new episode that loads a previously trained model. You can specify the episode and epoch to load from. By default, the last saved epoch is used.

```python
project.run_episode(
    'episode_6',
    parameters_update={'training': {'batch_size': 64}},
    load_episode='episode_2',
    load_epoch=100,
)
```

--------------------------------

### Initialize a new dlc2action Project via Terminal

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Use the dlc2action_init command to create a new project from the terminal, providing project details as arguments.

```bash
$ dlc2action_init --name project_name -d data_type -a annotation_type -dp path/to/data_folder -ap path/to/annotation_folder
```

--------------------------------

### Run a Single Training Episode

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Execute a single training episode using default project parameters. The results are saved under the specified episode name in the meta files.

```python
project.run_episode('episode_1')
```

--------------------------------

### Run Training Episode with Specific Parameters

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Loads specific parameters for training and runs the 'episode_best_params' episode. The `round_to_binary` parameter is useful for parameters like feature map count, rounding to the nearest power of two.

```python
project.run_episode(
    'episode_best_params',
    load_search='search_1',
    load_parameters=['data/overlap', 'general/ssl'],
)
```

--------------------------------

### Create a new dlc2action Project in Python

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Instantiate the Project class to create a new project, specifying essential details like name, data types, and paths.

```python
from dlc2action.project import Project

project = Project(
    'project_name',
    data_type='data_type',
    annotation_type='annotation_type',
    data_path='path/to/data/folder',
    annotation_path='path/to/annotation/folder',
)
```

--------------------------------

### Update Parameters and Run Episode with Best Parameters

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Loads specific parameters from a search result to update configuration files and runs a training episode using other parameters from the same search. The `round_to_binary` option can be used to round certain parameters to the nearest power of two.

```python
project.update_parameters(
    load_search='search_1',
    load_parameters=['training/lr', 'model/num_f_maps'],
    round_to_binary=['model/num_f_maps'],
)
project.run_episode(
    'episode_best_params',
    load_search='search_1',
    load_parameters=['data/overlap', 'general/ssl'],
)

```

--------------------------------

### List Episodes with Filters

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Provides a summary table of training episodes, with options to filter by episode names, display specific parameters, and apply value-based filters on parameters like recall and training time.

```python
project.list_episodes(
    episodes=['episode_1', 'episode_2', 'episode_3', 'episode_4'],
    display_parameters=['meta/time', 'results/recall', 'training/lr'],
    episode_filter='results/recall:>0.4,meta/training_time:>=00:30:00',
)
```

--------------------------------

### Run Hyperparameter Search

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Initiates a hyperparameter search using Optuna. Define the search space, optimization metric, number of trials, and other search parameters. Pruning can be enabled for efficiency.

```python
project.run_hyperparameter_search(
    search_space={
        "data/overlap": ("int", 10, 80),
        "general/ssl": (
            "categorical",
            [["contrastive"], ["contrastive", "masked_features"]]
        ),
        "training/lr": ("float_log": 1e-4, 1e-2),
        "model/num_f_maps": ("int_log", 8, 64),
    },
    metric="recall",
    n_trials=50,
    average=5,
    search_name="search_1",
    direction="maximize",
    prune=True,
)
```

--------------------------------

### List and Filter Training Episodes

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Lists training episodes with filtering options for episode names, displayed parameters, and parameter values. Filters can be combined for precise data retrieval.

```python
project.list_episodes(
    episodes=['episode_1', 'episode_2', 'episode_3', 'episode_4'],
    display_parameters=['meta/time', 'results/recall', 'training/lr'],
    episode_filter='results/recall::>0.4,meta/training_time::>=00:30:00',
)
```

--------------------------------

### Run Hyperparameter Search with dlc2action

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Initiates a hyperparameter search using optuna. Define the search space with parameter names and their types (int, float, categorical), specify the optimization metric, number of trials, and other search parameters like averaging, search name, direction, and pruning.

```python
project.run_hyperparameter_search(
    search_space={
        "data/overlap": ("int", 10, 80),
        "general/ssl": (
            "categorical",
            [["contrastive"], ["contrastive", "masked_features"]]
        ),
        "training/lr": ("float_log": 1e-4, 1e-2),
        "model/num_f_maps": ("int_log", 8, 64),
    },
    metric="recall",
    n_trials=50,
    average=5,
    search_name="search_1",
    direction="maximize",
    prune=True,
)

```

--------------------------------

### Continue a Previously Run Episode

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Resume training for an existing episode. All parameters and state dictionaries are loaded, allowing training to continue from where it stopped. The episode will be re-saved under the same name. Note that `num_epochs` specifies the new total number of epochs.

```python
project.continue_episode('episode_2', num_epochs=500)
```

--------------------------------

### Run Multiple Training Episodes Consecutively

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Train multiple episodes in a sequence. This can save time by reusing the Task instance across episodes.

```python
project.run_episodes(['episode_2', 'episode_3', 'episode_4'])
```

--------------------------------

### Fill dlc2action Project Parameters via Terminal

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Use the dlc2action_fill command to update project configuration parameters. The --all flag prompts for all parameters, otherwise it focuses on missing '???' fields.

```bash
$ dlc2action_fill --name project_name --all
```

--------------------------------

### Visualize Model Predictions

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Generates a prediction for a random sample and visualizes it against the ground truth. Customize parameters for different visualization needs.

```python
project.visualize_results('episode_1', load_epoch=50)
```

--------------------------------

### Plot Training Curves for Multiple Episodes

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Compares training or validation accuracy curves across multiple episodes. Specify the episodes and metrics to plot.

```python
project.plot_episodes(['episode_1', 'episode_2'], metrics=['accuracy'])
```

--------------------------------

### Run Prediction and Suggestion Tasks

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Executes prediction tasks to generate pickled dictionaries or suggestion files for the GUI. Supports specifying the episode, epoch, data subset, or custom data path.

```python
project.run_prediction('prediction_1', episode_name='episode_3', load_epoch=150, mode='test')
```

```python
project.run_suggestion(
    'suggestion_new_data',
    suggestion_episode='episode_4',
    suggestion_classes=['sleeping', 'eating'],
    suggestion_threshold=0.6,
    exclude_classes=['inactive'],
    data_path='/path/to/new_data_folder',
)
```

--------------------------------

### Update Parameters from Search Results

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Loads the best parameters from a completed hyperparameter search to update the current configuration.

```python
project.update_parameters(
    load_search='search_1',
```

--------------------------------

### List Best Parameters from a Search

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Retrieves and displays the best parameters found from a previously completed hyperparameter search. Specify the name of the search to query.

```python
project.list_best_parameters('search_1')

```

--------------------------------

### List Best Hyperparameter Search Results

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Retrieves and displays the best parameters found from a specified hyperparameter search.

```python
project.list_best_parameters('search_1')
```

--------------------------------

### Plot Training and Validation Recall Curves

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Plots training and validation recall curves for a specified episode. Useful for analyzing performance trends within a single episode.

```python
project.plot_episodes(['episode_3'], metrics=['recall'], modes=['train', 'val'])
```

--------------------------------

### Plot Combined Episode Curves

Source: https://amathislab.github.io/DLC2action/html_docs/dlc2action.html

Plots several episodes as a single curve, useful for comparing models where one loaded from another. Specify episodes and metrics.

```python
project.plot_episodes([['episode_2', 'episode_6'], 'episode_4'], metrics=['precision'])
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.