The transition to agriculture is regarded as a major turning point in human history. In the present contribution we propose to look at it through the lens of ethnographic data by means of a machine learning approach. More specifically, we analyse both the subsistence economies and the socioecological context of 1290 societies documented in the Ethnographic Atlas with a threefold purpose: (i) to better understand the variability and success of human economic choices; (ii) to assess the role of environmental settings in the configuration of the different subsistence economies; and (iii) to examine the relevance of fishing in the development of viable alternatives to cultivation. All data were extracted from the publicly available cross-cultural database D-PLACE. Our results suggest that not all subsistence combinations are viable, existing just a subset of successful economic choices that appear recurrently in specific ecological systems. The subsistence economies identified are classified as either primary or mixed economies in accordance with an information-entropy-based quantitative criterion that determines their degree of diversification. Remarkably, according to our results, mixed economies are not a marginal choice, as they constitute 25% of the cases in our data sample. In addition, fishing seems to be a key element in the configuration of mixed economies, as it is present across all of them.