FeatureTools
Powerful, mainly used in building automation engineering characteristics;
Three important: the entity, wherein the primitive, the DFS;
Entity: similar to a table; relationships between a plurality of entities can be constructed, similar to the association table; a plurality of set of entities, entity;
Primitive features: a similar manner to the process table fields, a characteristic of processing is called primitives, and may be custom signature motif, such as: the summation SUM (), the minimum value min (), the average avg ( ) Wait;
DFS: Similar encapsulated object can access the entity set, wherein the base element; wherein the primitives defined processing rule, applied to the entity, the entity statistics output;
E.g:
Entity / data: membership information, order information, product information
Entity Relationship / data relationship: Member and order-to-many, many orders and products;
Feature Units / statistical rules: Member of orders, the number of members of the commodity, the largest amount of orders members, average order amount of members, membership of a minimum amount of orders and so on;
1, the configuration of the field type of the entity and other information (may not be arranged, the DFS may infer the type automatically, but incorrect);
2, wherein the configuration information of primitives, such as: MODE (), MEAN (), SUM (), STD ();
3, DFS access configuration information, the entity according to the field type, wherein using the corresponding primitives calculated the result;
problem:
1, when too much data table, field too, the configuration involves a lot of work would be;
2, when the data is too large, based on python need to enable multi-process / multi-thread computing, the development of tuning workload;
github Address:
https://github.com/FeatureLabs/featuretools