-
Notifications
You must be signed in to change notification settings - Fork 0
Project Vision
This section outlines the core vision for the program, detailing various scenarios that reflect both its intended use and potential limitations. The following scenarios will guide the software design, its user interaction, and its evolution over time. These scenarios express the CREATOR’S INTENT, assist USERS in interpreting the program’s behavior, and offer insights to COLLABORATING PROGRAMMERS on future enhancements or adaptations.
User: Sarah, a data scientist, is working on an NLP project where she needs to annotate a large dataset with labels for training a text classification model. She opens the program and loads her dataset, which is in CSV format. The program automatically identifies the sample class type based on the configuration file, which specifies the TextClassificationSample class. Sarah is able to review the dataset, select specific columns for text and label, and run the annotation process without encountering any errors. The program allows her to save the annotated dataset in a new file format that matches the requirements for model training. Sarah also uses the program to randomly select a subset of samples for review, ensuring the quality of annotations before proceeding with model training. The program’s efficiency and ease of use significantly reduce Sarah’s time spent on manual labeling.
Creator’s Intent: This scenario demonstrates the program’s primary functionality of automating the annotation of datasets in a structured manner. It also reflects the seamless integration between dataset loading, sample creation, annotation, and output generation. This process is central to the tool’s purpose and should remain straightforward and error-free.
User: Mark, an AI researcher, is tasked with implementing a new sequence-to-sequence model for an NLP project. He wants to annotate data for a named entity recognition task using the program. Mark loads his dataset and chooses the SequenceToSequenceSample class, which is pre-configured for sequence labeling tasks. The program processes the dataset and automatically applies the correct sequence of labels to each input sample. Mark is able to review and save the annotated data, then export it in the format required by his model. The program's flexibility in adapting to different sample types and annotations, along with its intuitive interface, makes it easy for Mark to integrate it into his workflow.
Creator’s Intent: This scenario highlights the program's ability to support multiple types of sample annotations (e.g., text classification, sequence-to-sequence) and its adaptability to various user needs. The program is designed to allow researchers and developers to easily switch between different types of tasks with minimal configuration. This should remain a core feature of the program, supporting both flexibility and extensibility.
User: Emily, a machine learning engineer, has prepared a dataset for text classification, but she saved it in a non-standard format (e.g., a custom JSON format) that is not compatible with the program. She opens the program, but when she attempts to load the dataset, it fails to recognize the structure and throws an error. The error message is vague, and Emily struggles to understand the cause of the issue. After several attempts to reformat the dataset in CSV, she realizes that the program does not yet support custom dataset formats without proper preprocessing. Frustrated, Emily spends additional time manually converting the dataset into an accepted format before proceeding with the annotation.
Creator’s Intent: This scenario exposes a known limitation: the program currently supports a limited set of input formats (e.g., CSV, JSON) and may not handle custom formats or datasets with non-standard structures. While the program provides flexibility within predefined formats, users must ensure the dataset is properly formatted. The program’s design should account for expanding support for additional formats in future versions, but for now, users must be aware of these limitations.
User: Alex, a student working on a text classification project, uses the program to annotate a dataset of customer reviews. However, when he runs the annotation process, the program outputs incorrect labels for some of the reviews. After investigating, Alex realizes that the configuration file was not properly set up to match the specific class labels used in his dataset. The program did not provide a clear error or warning when the configuration was incorrect, leading to confusion about why the results were incorrect. Alex has to manually correct the configuration and re-run the annotation process to obtain the correct results.
Creator’s Intent: This scenario identifies the program’s potential to cause confusion if the configuration file is not correctly set up. A clear and intuitive error-handling mechanism, along with better user feedback, would help avoid such issues. This feedback should be incorporated into future updates to ensure that the user is informed of configuration errors in real time.
These scenarios guide the program’s design and intended use by clarifying the user experience, possible limitations, and areas for improvement. As development progresses, they will continue to serve as a reference to ensure that the software aligns with the creators’ vision and meets the users' needs effectively. Additionally, they offer valuable insights for future iterations, encouraging improvements and expanding capabilities.