Models and Model Training in Kudra
Last updated
Last updated
In Kudra, models are advanced AI algorithms designed to perform specific tasks such as extracting entities, identifying relationships, or classifying documents. These models can be trained within the platform using your data or by uploading pre-trained models for specialized tasks.
Click on the Cube Icon/Models:
Find and click on the cube icon to navigate to the Models page.
On the Models page, you will see either an empty page with options to create or upload a model, or a list of existing models. This page is divided into tabs that categorize the models by their types:
Tabs for Model Types
The Models page includes several tabs to help organize your models:
Text Annotation
Relation Annotation
Span Categorizer
Hugging Face
Classification
Processor
Each tab corresponds to a different type of model, allowing you to easily switch between them and manage your models effectively.
When a tab is selected and there are no models of that type associated with your account, you will see a message indicating that there's nothing to show, along with an option to upload a model.
These models enable Kudra users to automate complex data extraction and classification tasks, enhancing productivity and accuracy in document processing workflows.
Click on the Cube Icon:
Find and click on the cube icon to navigate to the Models page.
Click the "Create Model" Button:
On the Models page, click the "Create Model" button.
A popup will appear with the following fields to fill:
Project: Select the project this model will belong to.
Model Name: Enter a name for your model.
Model Category: Choose the type of model (e.g., Named Entity Recognition).
After filling in the fields, click "Create Model"
Click the "Upload Model" Button:
Navigate to the models page where you can upload a model.
Click the "Upload Model" button. Supported models are Spacy, BERT and LayoutLM
Fill in the Upload Form:
A popup will appear with fields to complete:
Project: Select the project this model will belong to.
Model Name: Enter a name for your model.
Model Category: Choose the type of model (e.g., Named Entity Recognition).
Pre-trained Model Type: Select the type of pre-trained model you are uploading.
Upload Section: Drag and drop the pre-trained model files or use the upload button to select files from your computer.
Review the details and click the "Upload" button to start the process.
By following these steps, you can successfully upload a pre-trained model in Kudra for use in your projects.
1 . Navigate to the Models Page:
Click on the Models option in the navigation menu.
2 . Select the Appropriate Tab:
Choose the tab that corresponds to the type of model you want to train.
3 . Locate the Model:
Find the model you want to train from the list.
4 . Open the Model Options:
Click on the three dots (ellipsis) button in the row of the model you want to train.
5 . Select the "Train" Option:
From the drop-down menu, click on the Train option.
6 . Configure the Model for Training:
A modal window will appear to configure the training settings for the model. Fill in the necessary details and start the training process.
The Model History section provides detailed information about the past actions and performance metrics associated with each model. This is crucial for tracking progress, understanding the efficacy of different models, and making data-driven decisions.
Key Columns in the Model History:
Workflow:
Displays the specific workflow in which the model was used.
Created At:
Indicates the date and time when the model was created. This is useful for keeping track of model versions and understanding the timeline of developments.
Type:
Specifies the type of model. This allows users to quickly identify the model's purpose.
Pre-Trained Model:
Shows whether the model was built from scratch or based on a pre-trained model. This information is crucial for understanding the model's foundation and initial capabilities.
P (Precision):
Reflects the precision score of the model, indicating how many of the model's positive predictions were actually correct.
What it Means: Precision measures how many of the items identified by the model are actually correct.
In Simple Terms: Imagine you are picking apples from a basket, but you accidentally pick some oranges too. Precision tells you how good you are at picking only apples without making mistakes. High precision means you picked mostly apples and very few oranges.
R (Recall):
Represents the recall score, showing how many of the actual positives were correctly identified by the model.
What it Means: Recall measures how good the model is at finding all the relevant information in your documents.
In Simple Terms: Imagine you are looking for all the apples in a basket of mixed fruits. If you find most of the apples, your recall is high. If you miss a lot of apples, your recall is low.
F (F1 Score):
Combines precision and recall into a single metric, providing a balanced measure of the model's accuracy. The F1 Score is particularly useful when you need to balance precision and recall.
What it Means: The F1 score combines recall and precision into a single number to give a balanced view of the model's performance.
In Simple Terms: Think of the F1 score as a way to see how well the model balances between finding all the apples (recall) and not picking up other fruits by mistake (precision). A high F1 score means the model is good at finding the right information without too many mistakes.
Training/Validation:
Provides information on the training and validation datasets used for the model. This helps in understanding the data context and the robustness of the model's training process.
Entity Score:
Displays the entity score, which measures the model's accuracy in extracting specific entities. This is crucial for models focused on entity extraction tasks.
Importance of Model History:
Track Performance: Monitor how different versions of models have performed over time.
Data-Driven Decisions: Make informed decisions based on the historical performance metrics.
Identify Trends: Spot trends and patterns in model performance to guide future improvements.
Audit Trail: Maintain an audit trail for compliance and review purposes.
Keeping a detailed history of model performance and changes is essential for continuous improvement and ensuring the reliability of the AI models within Kudra.
