san_francisco_example.py

import argparse
import os
from auto_mm_bench.datasets_with_image import dataset_with_image_registry, create_dataset
from autogluon.core.features.feature_metadata import FeatureMetadata
from autogluon.tabular import TabularPredictor

BASELINE_HPARAMS = {
    'FASTAI': {},
}

train_dataset = create_dataset('san_francisco_airbnb', 'train')

single_image_train_df = train_dataset.get_df_with_single_image()
single_image_train_df = single_image_train_df[list(set(train_dataset.feature_columns + train_dataset.label_columns[:1]) - set(train_dataset.image_columns[1:]))]

feature_metadata = FeatureMetadata.from_df(single_image_train_df).add_special_types(
    {train_dataset.image_columns[0]: ['image_path']})
eval_metric = train_dataset.metric

predictor = TabularPredictor(label=train_dataset.label_columns[0],
                             path='san_francisco_airbnb_tabular',
                             problem_type=train_dataset.problem_type, # specify problem_type
                             eval_metric=eval_metric)
predictor.fit(
    train_data=single_image_train_df,
    hyperparameters=BASELINE_HPARAMS,
    feature_metadata=feature_metadata
)

test_dataset = create_dataset('san_francisco_airbnb', 'test')
predictor.leaderboard(test_dataset.get_df_with_single_image())