torch_frame

The semantic type of a column.

A semantic type denotes the semantic meaning of a column, and denotes how columns are encoded into an embedding space within tabular deep learning models:

import torch_frame

stype = torch_frame.numerical  # Numerical columns
stype = torch_frame.categorical  # Categorical columns
...

numerical: Numerical columns.

categorical: Categorical columns.

text_embedded: Pre-computed embeddings of text columns.

text_tokenized: Tokenized text columns for finetuning.

multicategorical: Multicategorical columns.

sequence_numerical: Sequence of numerical values.

embedding: Embedding columns.

timestamp: Timestamp columns.

image_embedded: Pre-computed embeddings of image columns.

class Metric(value)[source]

The metric.

ACCURACY: accuracy

ROCAUC: rocauc

RMSE: rmse

MAE: mae

class TaskType(value)[source]

The type of the task.

REGRESSION: Regression task.

MULTICLASS_CLASSIFICATION: Multi-class classification task.

BINARY_CLASSIFICATION: Binary classification task.

class NAStrategy(value)[source]

Strategy for dealing with NaN values in columns.

MEAN: Replaces NaN values with the mean of a torch_frame.numerical column.

ZEROS: Replaces NaN values with zeros in a torch_frame.numerical column.

MOST_FREQUENT: Replaces NaN values with the most frequent category of a torch_frame.categorical column.