Skip to main content
Version: Next

VAD Parameter

Updated 2024.06.11

experimental_plan.yaml Description​

To apply AI Contents to the data you have, you need to enter information about the data and the Content features to be used in the experimental_plan.yaml file. When you install AI Contents in the solution folder, you can check the experimental_plan.yaml file provided for each content under the solution folder. Enter the 'data information' in this yaml file and modify/add the 'user arguments' provided for each asset. When you run ALO with the modified user arguments, you can generate a data analysis model with the desired settings.

experimental_plan.yaml Structure​

experimental_plan.yaml contains various setting values necessary to run ALO. By modifying the 'data path' and 'user arguments' parts, you can use AI Contents immediately.

Enter Data Path (external_path)​

  • The parameters of external_path are used to specify the path to load or save files. If save_train_artifacts_path and save_inference_artifacts_path are not entered, the modeling outputs are saved in the default directories train_artifacts and inference_artifacts.
external_path:
    - load_train_data_path: ./solution/sample_data/train
    - load_inference_data_path:  ./solution/sample_data/test
    - save_train_artifacts_path:
    - save_inference_artifacts_path:
Parameter NameDEFAULTDescription and Options
load_train_data_path./sample_data/train/Enter the folder path where the training data is located (do not enter the csv file name).
load_inference_data_path./sample_data/test/Enter the folder path where the inference data is located (do not enter the csv file name).

User Parameters (user_parameters)​

  • Below user_parameters, step represents the asset name. step: input below indicates the input asset stage.
  • args represents the user arguments of the input asset (step: input). User arguments are parameters related to data analysis provided for each asset. Please refer to the explanation below for details.
user_parameters:
	- train_pipeline:
		- step: input
		  args: 
			- file_type
			...
		  ui_args:
			...

User Arguments Explanation​

What are User Arguments?​

User arguments are parameters for configuring the operation of each asset in the experimental_plan.yaml under the args of each asset step. They are used to customize various functionalities of the AI Contents pipeline for your data. Users can change or add user arguments based on the following guide to tailor the modeling to their data. User arguments are divided into "Required arguments," which are predefined in the experimental_plan.yaml, and "Custom arguments," which users add based on the provided guide.

Required Arguments​

  • Required arguments are the basic arguments shown directly in the experimental_plan.yaml. Most required arguments have default values built in. For arguments with default values, the user does not need to set a separate value, as it will operate with the default value.
  • Among the required arguments in the experimental_plan.yaml, data-related arguments must be set by the user. (e.g., path_column, y_column)

Custom Arguments​

  • Custom arguments are not listed in the experimental_plan.yaml but are functionalities provided by the asset, which users can add to the experimental_plan.yaml for use. They are added to each asset's 'args' section.

The VAD pipeline consists of Input - Readiness - Modeling (train/inference) - Output assets, and user arguments are configured differently for each asset according to its functionality. Start by using the required user arguments listed in the experimental_plan.yaml and then add more user arguments to tailor the VAD model to your data!

User Arguments Summary​

Here's a summary of user arguments for VAD. Click on the 'Argument Name' to navigate to the detailed explanation of the argument.

Default​

  • The 'Default' section indicates the default value of the user argument.
  • If there's no default value, it's represented as '-'.
  • If there's a logic for the default value, it's indicated as 'Refer to Explanation'. Click on 'Argument Name' to see the detailed explanation.

ui_args​

  • The 'ui_args' in the table below indicates whether the argument supports the ui_args feature in AI Conductor's UI for changing argument values.
  • O: If the argument name is entered under the ui_args in the experimental_plan.yaml, it can be changed in the AI Conductor UI.
  • X: Does not support the ui_args feature.
  • For more information on ui_args, refer to the following guide: Write UI Parameter
  • In the FCST experimental_plan.yaml, all user arguments that can be ui_args are pre-filled under ui_args_detail.

User Setting Requirement​

  • The 'User Setting Requirement' in the table below indicates whether the user needs to confirm and change the user arguments to run AI Contents.
  • O: These are typically arguments related to tasks or data that the user needs to input before modeling.
  • X: Modeling will proceed with default values if the user does not change the values.
Asset NameArgument TypeArgument NameDefaultDescriptionUser Setting Requirementui_args
InputCustomfile_typecsvEnter the file extension of the input data.OX
InputCustomencodingutf-8Enter the encoding type of the input data.XX
ReadinessCustomok_class-Enter the class representing 'ok' in the y_column.XX
ReadinessCustomtrain_validate_column-Enter the column that distinguishes between train and validation. This column should consist of two parameters: train and valid.XX
ReadinessCustomvalidation_split_rate0.1If train_validate_column does not exist, this rate generates validation from the input train data.XX
ReadinessCustomthreshold_methodF1Selects the method for determining OK and NG during validation. Possible values: F1, Percentile (automatically selected depending on the validation data size).XX
ReadinessCustomnum_minor_threshold5If the number of OK and NG images does not exceed this, threshold_method is automatically selected as Percentile.XX
ReadinessCustomratio_minor_threshold0.1If the ratio of OK and NG images does not exceed this, threshold_method is automatically selected as Percentile.XX
TrainRequiredmodel_namefastflowSelects the model to be used for VAD. Possible values: fastflow, patchcoreOO
TrainCustomexperiment_seed42Determines the experiment seed in the PyTorch environment.XX
TrainCustomimg_size[256,256]Sets the image size for resizing during image training. If only numbers are entered, the aspect ratio of the original image is maintained during resize (the shorter side becomes the resize size, maintaining the aspect ratio).XX
TrainCustombatch_size4Sets the batch size during training and validation.XX
TrainCustommax_epochs15Sets the maximum number of epochs for training.XX
TrainCustomacceleratorcpuSelects whether to run based on CPU or GPU. GPU is recommended in environments where it's available.XX
TrainCustommonitor_metricimage_AUROCSelects the criterion for saving the best model. If threshold_method is Percentile, loss is automatically selected. Possible values: loss, image_AUROC, image_F1ScoreXX
TrainCustomsave_validation_heatmapTrueSelects whether to save prediction heatmaps for the validation dataset. (Only saves for ok-ng, ng-ok, and ng-ng cases)XX
TrainCustompercentile0.8Selects the percentile of the anomaly score of the validation dataset to judge as NG when threshold_method is Percentile.XX
TrainCustomaugmentation_list[]List of transformations applied for random augmentation. Can include rotation, brightness, contrast, saturation, hue, and blur.XX
TrainCustomaugmentation_choice3Number of times transformations are applied for random augmentation. Possible values: non-negative integersXX
TrainCustomrotation_angle10Maximum angle for rotation in random augmentation. If set to 10, transformation is performed between -10 and 10. Possible values: non-negative integers between 0 and 180XX
TrainCustombrightness_degree0.3Degree of brightness adjustment in random augmentation. Sets maximum/minimum brightness levels. Possible values: floating-point numbers between 0 and 1XX
TrainCustomcontrast_degree0.3Degree of contrast adjustment in random augmentation. Possible values: floating-point numbers between 0 and 1XX
TrainCustomsaturation_degree0.3Degree of saturation adjustment in random augmentation. Possible values: floating-point numbers between 0 and 1XX
TrainCustomhue_degree0.3Degree of hue adjustment in random augmentation. Possible values: floating-point numbers between 0 and 1XX
TrainCustomblur_kernel_size5Maximum kernel size for blur in random augmentation. Possible values: integers between 0 and 255XX
TrainCustommodel_parameters{"fastflow_backborn": 'resnet18', "fastflow_flow_steps": 8, "patchcore_backborn": 'wide_resnet50_2', "patchcore_coreset_sampling_ratio": 0.1, "patchcore_layers": ["layer2", "layer3"]}Parameters related to model training. If not set, the model is trained with default parameters. For details, refer to the parameter description below.XX
InferenceCustominference_threshold0.5Threshold of anomaly score for being judged as abnormal.XX
InferenceCustomsave_anomaly_mapsFalseWhether to save XAI image results.XX

User Arguments Detailed Description​

Input asset​

file_type​

Enter the file extension of the input data.

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • csv (default)
  • Usage
    • file_type: csv
  • ui_args: X

encoding​

Enter the encoding type of the input data.

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • utf-8 (default)
  • Usage
    • encoding: utf-8
  • ui_args: X

Readiness asset​

ok_class​

Enter the class representing 'ok' in the y_column. If not entered, ok_class is entered as the name that occupies the most in the training dataset.

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • '' (default)
  • Usage
    • ok_class: good
  • ui_args: O

train_validate_column​

Enter the column that distinguishes between train and validation. This column should consist of two parameters: train and valid.

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • '' (default)
  • Usage
    • train_validate_column: phase
  • ui_args: X

validation_split_rate​

If train_validate_column does not exist, this rate generates validation from the input train data.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.1 (default)
  • Usage
    • validation_split_rate: 0.1
  • ui_args: X

threshold_method​

Selects the method for determining OK and NG during validation. Possible values: F1, Percentile (automatically selected based on the number of validation data).

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • F1 (default)
    • Percentile
  • Usage
    • threshold_method: Percentile
  • ui_args: X

num_minor_threshold​

If the number of OK and NG images does not exceed this, the threshold_method is automatically selected as Percentile.

  • Argument type: Custom
  • Input type
    • integer
  • Possible values
    • 5 (default)
  • Usage
    • num_minor_threshold: 5
  • ui_args: X

ratio_minor_threshold​

If the ratio of OK and NG images does not exceed this, the threshold_method is automatically selected as Percentile.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.1 (default)
  • Usage
    • ratio_minor_threshold: 0.1
  • ui_args: X

Train asset​

model_name​

Selects the model to be used for VAD. Possible values: fastflow, patchcore

  • Argument type: Required
  • Input type
    • string
  • Possible values
    • fastflow (default)
    • patchcore
  • Usage
    • model_name: patchcore
  • ui_args: O

img_size  ​

Sets the image size for training. If only numbers are entered, the aspect ratio of the original image is maintained during resizing.

  • Argument type: Custom
  • Input type
    • integer
  • Possible values
    • [256,256] (default)
    • 256
  • Usage
    • img_size: [256,256]
  • ui_args: X

batch_size​

Sets the batch size for training and validation.

  • Argument type: Custom
  • Input type
    • integer
  • Possible values
    • 4 (default)
  • Usage
    • batch_size: 32
  • ui_args: X

experiment_seed​

Determines the experiment seed in the PyTorch environment.

  • Argument type: Custom
  • Input type
    • integer
  • Possible values
    • 42 (default)
  • Usage
    • experiment_seed: 42
  • ui_args: X

max_epochs​

Sets the maximum number of epochs for training.

  • Argument type: Custom
  • Input type
    • integer
  • Possible values
    • 15 (default)
  • Usage
    • max_epochs: 15
  • ui_args: X

accelerator​

Selects whether to run based on CPU or GPU.

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • cpu (default)
    • gpu
  • Usage
    • accelerator: gpu
  • ui_args: X

monitor_metric​

Selects the criterion for saving the best model. Possible values: loss, image_AUROC, image_F1Score

  • Argument type: Custom
  • Input type
    • string
  • Possible values
    • image_AUROC (default)
    • loss
    • image_F1Score
  • Usage
    • monitor_metric: loss
  • ui_args: X

save_validation_heatmap​

Selects whether to save prediction heatmaps for the validation dataset.

  • Argument type: Custom
  • Input type
    • bool
  • Possible values
    • True (default)
    • False
  • Usage
    • save_validation_heatmap: True
  • ui_args: X

percentile​

Selects the percentile of anomaly score in the validation dataset to determine NG when threshold_method is Percentile.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.8 (default)
  • Usage
    • percentile: 0.8
  • ui_args: X

augmentation_list​

List of transformations applied for random augmentation. Multiple options can be used: rotation, brightness, contrast, saturation, hue, blur.

  • Argument type: Custom
  • Input type
    • list
  • Possible values
    • [] (default)
    • rotation, brightness, contrast, saturation, hue, blur
  • Usage
    • augmentation_list: [rotation, brightness, contrast, saturation, hue, blur]
  • ui_args: X

augmentation_choice​

Number of times transformations are applied for random augmentation.

  • Argument type: Custom
  • Input type
    • int
  • Possible values
    • 3 (default)
  • Usage
    • augmentation_choice: 2
  • ui_args: X

rotation_angle​

Maximum angle for rotation in random augmentation. If set to 10, transformation is performed between -10 and 10.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 10 (default)
  • Usage
    • rotation_angle: 30
  • ui_args: X

brightness_degree​

Degree of brightness adjustment in random augmentation. Sets maximum/minimum brightness levels.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.3 (default)
  • Usage
    • brightness_degree: 0.5
  • ui_args: X

contrast_degree​

Degree of contrast adjustment in random augmentation.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.3 (default)
  • Usage
    • contrast_degree: 0.5
  • ui_args: X

saturation_degree​

Degree of saturation adjustment in random augmentation.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.3 (default)
  • Usage
    • saturation_degree: 0.5
  • ui_args: X

hue_degree​

Degree of hue adjustment in random augmentation.

  • Argument type: Custom
  • Input type
    • float
  • Possible values
    • 0.3 (default)
  • Usage
    • hue_degree: 0.5
  • ui_args: X

blur_kernel_size​

Maximum kernel size for blur in random augmentation.

  • Argument type: Custom
  • Input type
    • int
  • Possible values
    • 15 (default)
  • Usage
    • blur_kernel_size: 25
  • ui_args: X

model_parameters​

Parameters related to model training. If not set, the model is trained with default parameters. For details, refer to the parameter description below.

  • Argument type: Custom
  • Input type
    • dictionary
  • Possible values
    • {"fastflow_backborn": 'resnet18', "fastflow_flow_steps": 8, "patchcore_backborn": 'wide_resnet50_2', "patchcore_coreset_sampling_ratio": 0.1, "patchcore_layers": ["layer2", "layer3"]} (default)
    • fastflow_backborn: "resnet18", "wide_resnet50_2", "cait_m48_448", "deit_base_distilled_patch16_384"
    • fastflow_flow_steps: int
    • patchcore_backborn: any model available in the timm package, but it should interact with the names of layers in patchcore_layers.
    • patchcore_coreset_sampling_ratio: float between 0 and 1
    • patchcore_layers: list of the names of selected backborn's layers
  • Usage
    • model_parameters: {"fastflow_backborn": 'resnet18', "fastflow_flow_steps": 8, "patchcore_backborn": 'wide_resnet50_2', "patchcore_coreset_sampling_ratio": 0.1, "patchcore_layers": ["layer2", "layer3"]}
  • ui_args: X

Inference asset​

inference_threshold​

Threshold of anomaly score for being judged as abnormal.

  • Argument type: Required
  • Input type
    • float
  • Possible values
    • 0.5 (default)
  • Usage
    • inference_threshold: 0.5
  • ui_args: X

save_anomaly_maps​

Whether to save XAI image results.

  • Argument type: Custom
  • Input type
    • bool
  • Possible values
    • False (default)
    • True
  • Usage
    • save_anomaly_maps: True
  • ui_args: X

***  

VAD Version: 1.0.0