MADlib
1.1 A newer version is available
User Documentation
|
SQL functions for low-rank matrix factorization. More...
Go to the source code of this file.
Functions | |
integer | lmf_igd_run (varchar rel_output, regclass rel_source, varchar col_row, varchar col_column, varchar col_value, integer row_dim="SELECT max(col_row) FROM rel_source", integer column_dim="SELECT max(col_col) FROM rel_source", integer max_rank=20, float8 stepsize=0.01, float8 scale_factor=0.1, integer num_iterations=10, float8 tolerance=0.0001) |
Low-rank matrix factorization of a incomplete matrix into two factors. More... | |
Definition in file lmf.sql_in.
integer lmf_igd_run | ( | varchar | rel_output, |
regclass | rel_source, | ||
varchar | col_row, | ||
varchar | col_column, | ||
varchar | col_value, | ||
integer | row_dim = "SELECT max(col_row) FROM rel_source" , |
||
integer | column_dim = "SELECT max(col_col) FROM rel_source" , |
||
integer | max_rank = 20 , |
||
float8 | stepsize = 0.01 , |
||
float8 | scale_factor = 0.1 , |
||
integer | num_iterations = 10 , |
||
float8 | tolerance = 0.0001 |
||
) |
This function takes as input the table representation of a incomplete matrix in the sparse (i, j, value) format and decomposes it into the specified set of most significant features of matrices of U and V matrix. The input matrix is expected to have dimension [1:row_dim][1:column_dim], but in sparse format.
rel_output | Name of the table that the factors will be appended to |
rel_source | Name of the table/view with the source data |
col_row | Name of the column containing cell row number |
col_column | Name of the column containing cell column number |
col_value | Name of the column containing cell value |
row_dim | Maximum number of rows of input |
column_dim | Maximum number of columns of input |
max_rank | Rank of desired approximation |
stepsize | Hyper-parameter that decides how aggressive that the gradient steps are |
scale_factor | Hyper-parameter that decides scale of initial factors |
num_iterations | Maximum number if iterations to perform regardless of convergence |
tolerance | Acceptable level of error in convergence. |
Definition at line 278 of file lmf.sql_in.