UI Configuration

You can pass arguments to the run command to configure the Texture interface.

• data: pd.DataFrame: The dataframe to parse and visualize.
• schema: DatasetSchema: a dataset schema (calculated automatically if none provided)
• load_tables: Dict[str, pd.DataFrame]: A dictionary of tables to load.
• create_new_embedding_func: A function that takes a string and returns a vector embedding

If only data is provided, Texture will automatically calculate the schema and load tables. The schema and tables can also be calculated manually by calling the preprocess function then passed to run:

# automatically calculate schema and tables
texture.run(df)

# OR manually preprocess
schema, load_tables = texture.preprocess(df)

DatasetSchema

The schema is a DatasetSchema object that describes the columns, types, and tables in the dataset. These column definitions are used to determine how the data will be visualized in the interface. Each schema has the following fields:

• name: The name of the dataset.
• columns: A list of Column objects that describe the columns in the dataset.
• primary_key: A Column object that describes the primary key of the dataset.
• has_embeddings: Whether the dataset has embeddings or not. If so, must contain a column named "vector" with embedding arrays
• has_projection: Whether the dataset has projections or not. If so, must contain columns named "umap_x" and "umap_y" with 2d projection coordinates.

Column

Column definition contains:

• name: The name of the column as it appears in the dataframe
• type: The type of the column. Options are text, number, categorical, and boolean.
• derivedSchema: A DerivedSchema object that describes how the column is derived from another table. Options are:
  • is_segment: Whether the column is a segment (e.g. word, phrase) or not (e.g. author, keyword).
  • table_name: The name of the table the column is in.
  • derived_from: The column in the main table that the column is derived from.
  • derived_how: How the column is derived from the derived table.

Embedding Functions

To support similarity search, Texture allows you to pass a function that will create a new embedding for a given string. This allows you to create embeddings on the fly for query strings. We recommend pre-computing embeddings for your data and passing them to Texture in the vector column and then using the same model in this function to create embeddings for query strings.

For example, the following code creates a new embedding for a given string using the sentence-transformers library:

def get_embedding(value: str):
    import sentence_transformers
    model = sentence_transformers.SentenceTransformer("all-mpnet-base-v2")
    e = model.encode(value)
    return e

Example Usage

See the demos for examples of how to configure the schema, columns, and embedding functions.