Description
Fit the selected model independently by group using one target vector and the
configured predictor fields. The fitted target is written to the output;
optional diagnostics include fit error and model parameters.
Synopsis
sgfit <stdin >stdout model=linear model.nf=5 match.mf=0 model.qc=NULL optim=mse optim.driver=steepgrad optim.niter=10Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
in |
ifile |
stdin
|
input dataset: 1. the first vector is the target (dependent variable) 2. the remaining vectors are predictors (independent variables) |
out |
ofile |
stdout
|
output dataset with the first vector replaced by fitted values and a fiterr field added |
model |
enum |
linear
|
fitting model: linear: y = a[0] + sum_i=1:n a[i]*x[i] match: y[i] = sum_k=-mf:nf a[k]*x[i-k] poly: y = sum_k=0:nf a[k]*x^k |
model.nf |
int |
5
|
causal filter order for match mode, or polynomial order for poly mode |
match.mf |
int |
0
|
number of non-causal matching-filter coefficients |
model.qc |
ofile |
NULL
|
optional output dataset containing fitted model parameters |
optim |
enum |
mse
|
fitting objective: mae: minimize absolute error for Laplacian noise mse: minimize squared error for Gaussian noise mme: minimize maximal error m4e: minimize fourth-power error as an approximation to mme |
optim.driver |
enum |
steepgrad
|
optimization algorithm |
optim.niter |
int |
10
|
number of optimization iterations |
Input and output
Input
input dataset: 1. the first vector is the target (dependent variable) 2. the remaining vectors are predictors (independent variables)
Output
output dataset with the first vector replaced by fitted values and a fiterr field added