The Shakespeare and Company Project makes three datasets available to download in CSV and JSON formats. The datasets provide information about lending library members; the books that circulated in the lending library; and lending library events, including borrows, purchases, memberships, and renewals. The datasets may be used individually or in combination site URLs are consistent identifiers across all three. The DOIs for each dataset are as follows: Members (https://doi.org/10.34770/ht30-g395); Books (https://doi.org/10.34770/g467-3w07); Events (https://doi.org/10.34770/2r93-0t85).
The Molino suite contains 75,000 galaxy mock catalogs designed to quantify the information content of any cosmological observable for a redshift-space galaxy sample. They are constructed from the Quijote N-body simulations (Villaescusa-Navarro et al. 2020) using the standard Zheng et al. (2007) Halo Occupation Distribution (HOD) model. The fiducial HOD parameters are based on the SDSS high luminosity samples. The suite contains 15,000 mocks at the fiducial cosmology and HOD parameters for covariance matrix estimation. It also includes (500 N-body realizations) x (5 HOD realizations)=2,500 mocks at 24 other parameter values to estimate the derivative of the observable with respect to six cosmological parameters (Omega_m, Omega_b, h, n_s, sigma_8, and M_nu) and five HOD parameters (logMmin, sigma_logM, log M_0, alpha, and log M_1). Using the covariance matrix and derivatives calculated from Molino, one can derive Fisher matrix forecasts on the cosmological parameters marginalized over HOD parameters.
Extrapolation -- the ability to make inferences that go beyond the scope of one's experiences -- is a hallmark of human intelligence. By contrast, the generalization exhibited by contemporary neural network algorithms is largely limited to interpolation between data points in their training corpora. In this paper, we consider the challenge of learning representations that support extrapolation. We introduce a novel visual analogy benchmark that allows the graded evaluation of extrapolation as a function of distance from the convex domain defined by the training data. We also introduce a simple technique, context normalization, that encourages representations that emphasize the relations between objects. We find that this technique enables a significant improvement in the ability to extrapolate, considerably outperforming a number of competitive techniques.