Stratifying Samples

This python file uses the Kmeans algorithm to stratify samples of a database. Its use is simple, you only need to put the columns where you want realizes the stratification, the total of return samples and the sizes of each sample.

The stratification is based on distribution of clusters. For example, if we have 20% of entries in the cluster A, the stratified samples will have nearly 20% of entries of the cluster A.

For the tuning of cluster' numbers we use the Elbow Method. The algorithm also realizes the automatic tranformation of categorical and datetime data to integer numbers and also normalization of data.

Normalization

For the normalization we use the Sklearn function, StandardScaler. The data returned of this function will have mean equals to 0 and standart deviation proximally to 1.

Algorithm Parameters

To generate the stratified samples you need to following this steps:

1. Assuming that you install the dependencies and this file, you need to create a new object Estratificacao, with the following parameters:
   • qtd_base (int) : Quantity of samples.
   • tamanho_segmentacao (List) : Length of each sample.
   • percentage (bool Optional) : True if the legth is in percentage values and False otherwise.
   • normalize (bool Optional) : True if the data need to be normalized and False otherwise.
   • min_clusters (int Optional) : Minimum number of clusters in the Elbow Method.
   • max_clusters (int Optional) : Maximum number of clusters in the Elbow Method.
2. With the object Estratification, you only need to call the function estratificar with the following parameters:
   • data (DataFrame) : The database that you want stratify samples.
   • colunas_segmentacao (List) : Columns used to stratify.
   • atributo_diferenciacao (str) : Columns with a different values for each entry.
   • colunas_retornadas (List) : Columns returned in the stratified samples.