Settings page

The settings page contains many tabs with moderate to high impacts on your project. We describe them on this page.

Parameters

This page displays the parameters of the project and allows editing or deleting the project.

• Change parameters to edit some parameters:
  • Project name: see Project Creation page
  • Text column(s): see Project Creation page
  • Language of the corpus: see Project Creation page
  • Column(s) for contextual information: see Project Creation page
  • Add N elements in the train set: Pick new elements from the dataset uploaded. This action ignores stratification.
• Delete project to delete the project, annotations and models. There is no going back.

Features

Displays the available features and their parameters.

• Add a new feature to compute new features (more details on the available features modes).
  • if using embeddings:
    • Feature name
    • Model to use
    • Context window size: The number of token per entry. After tokenization each input is truncated/padded to match this size.
  • if using fasttext (See the Features concept):
    • Feature name
    • Model to use
  • if using dfm: Using CountVectorizer
    • Feature name
    • TF-IDF: if set to True, applies TF-IDF.
    • n-grams: the size of n-grams to account for.
    • min term freq: the minimum frequency for an n-gram to be included in the vocabulary.
    • max term freq: the maximum frequency for an n-gram to be included in the vocabulary.
    • Norm: if set to True, normalize the features at the end
    • Log: if set to True, apply logarithm function to the features.
  • if using regex: creates a boolean feature indicating whether the regex query is found in the text input
    • Feature name
    • Regex query
  • if using dataset: imports from a column in the original dataset
    • Column to use
    • Type of the feature: either numerical or categorical.

Features can be downloaded from the Export page.

Import

The import tab allows you to import data for the current process outside of the Project Creation page.

• Import annotations: import the annotations for the text inputs in the train dataset. The IDs must match. If the labels differ, new labels will be created. If annotations already exist, they will be overwritten.
  • File to upload: Parquet, CSV or XLSX file, limit defined by the administrator.
  • Column for ID: column from the loaded dataset that must match the external IDs from the current text inputs.
  • Column for annotations: column from the loaded dataset that contains the annotations to import.
  • Import annotations to finalise the importation.
• Validation set and Test set: drop or import the validation or test set.
  • if a validation/test set exists,
    • Drop Test/Validation set: delete the set as well as the features. They will be re-computed.
  • if the validation/test set does not exist
    • File to upload: Parquet, CSV or XLSX file, limit defined by the administrator.
    • ID column: index to identify each element in the treatment ; either from an existing column or row number. If the chosen index is not unique, an internal index is created different from the external index.
    • Text column(s): The column(s) to use as text input. If several columns are selected, content of the columns will be concatenated with two linebreaks.
    • Column(s) for existing annotations: Load existing annotations. If labels are not found in the current scheme, they will be ignored.
    • Number of elements: Number of rows