Table 1. Flags used in Careless .

Flag	Default	Description
--test-fraction	0.0	Fraction of the data held out for cross-validation
--merge-half-datasets	—	If True, merge pairs of random half data sets separately after training. Generates output files for estimating CC1/2.
--half-dataset-repeats	1	Number of half-data-set merging repeats. Used to estimate the uncertainty of CC1/2.
-d, --dmin	None	Set maximum resolution in Å. By default, merge reflections to the maximum input resolution.
--separate-files	False	Create an individual output file for each input MTZ. If True, data are scaled together and merged separately.
--studentt-likelihood-dof	None	Invoke Student’s t-distributed error model and set its degrees of freedom
--refine-uncertainties	—	Use the model in Evans (2011 ▸) to adjust the
--positional-encoding-frequencies	4	Number of positional encoding (Zhong et al., 2019 ▸; Mildenhall et al., 2020 ▸) frequencies for metadata. Use with mlp-width to limit memory use. Default encodes all columns. For sepcific columns, see --positional-encoding-keys.
--positional-encoding-keys	None	Provide comma-separated metadata keys for positional encoding
--anomalous	—	Merge Friedel mates separately
--wilson-prior-b	None	Learn reflections on a specific global Wilson scale. The default Wilson prior is flat across all resolutions.
--mlp-layers	20	Number of neural network layers
--mlp-width	None	Width of hidden layers in the neural net; defaults to the dimensionality of the metadata array
--image-layers	0	Add additional layers with local, image-specific parameters
--disable-image-scales	—	Do not learn a scalar, per-image scale parameter
-w, --wavelength-key	Wavelength	For polychromatic data, the MTZ column name corresponding to reflections’ peak wavelength