# psiz-datasets ## Purpose The OSF `psiz-datasets` repository is the hosting home for human behavior data files and pre-trained models associated with the PsiZ ecosystem. Data files are hosted in a relatively unprocessed format to allow users flexibility and mimimal coupling with a particular programming framework. If requiring pre-formatted data, please see the `psiz-datasets` python package (https://github.com/psiz-org/psiz-datasets), which will load the data as a TensorFlow `tf.data.Dataset` object. ## Getting Started Please see the PsiZ documentation (https://psiz.readthedocs.io/en/latest/) for guidance on using the datasets. ## Organization Distinct datasets are organized into OSF *components*. Within each component there is at least one directory called `data`, which contains the collected human behavior. The `data` directory may contain multiple versions that map to package compatibility. ## Old Versions Previously, datasets were coupled with the PsiZ python package (https://github.com/roads/psiz). Begining in psiz v0.8.0, dasets were moved to a separate `psiz-datasets` GitHub repository. Data was previously hosted in the following format: 1. The `obs.hdf5` file contains trials of human similarity judgments. This file may have a slightly different name (e.g., `obs-118.hdf5`). In this file, stimuli are recorded using unique IDs (not filenames). 2. The `catalog.hdf5` is primarily a mapping between stimuli IDs and file names.