In [2]:
[global]
# The "true" sparse regression coefficient
parameter: beta = [3, 1.5, 0, 0, 2, 0, 0, 0]
In [3]:
# Simulate sparse data-sets
[lasso_1, ridge_1]
depends: R_library("MASS>=7.3")
# training and testing samples
parameter: N = (40, 200)
parameter: rstd = 3
parameter: replicate = [x+1 for x in range(5)]
source='regression_modules/simulate.R'
input: None, for_each = ['replicate']
output: train = f"data_{_replicate}.train.csv", test = f"data_{_replicate}.test.csv"
bash: expand = True
Rscript {source} seed={_replicate} N="c({paths(N):,})" b="c({paths(beta):,})" rstd={rstd} oftrain="'{_output[0]}'" oftest="'{_output[1]}'"
In [4]:
# Ridge regression model implemented in R
# Build predictor via cross-validation and make prediction
[ridge_2]
depends: R_library("glmnet>=2.0")
parameter: nfolds = 5
source='regression_modules/ridge.R'
output: pred = f"{_input[0]:nn}.ridge.predicted.csv", coef = f"{_input[0]:nn}.ridge.coef.csv"
bash: expand = True
Rscript {source} train="'{_input[0]}'" test="'{_input[1]}'" nfolds={nfolds} ofpred="'{_output[0]}'" ofcoef="'{_output[1]}'"
In [5]:
# LASSO model implemented in Python
# Build predictor via cross-validation and make prediction
[lasso_2]
depends: Py_Module("sklearn>=0.18.1"), Py_Module("numpy>=1.6.1"), Py_Module("scipy>=0.9")
parameter: nfolds = 5
source='regression_modules/lasso.py'
output: pred = f"{_input[0]:nn}.lasso.predicted.csv", coef = f"{_input[0]:nn}.lasso.coef.csv"
bash: expand = True
python {source} {_input[0]} {_input[1]} {nfolds} {_output[0]} {_output[1]}
In [6]:
# Evaluate predictors by calculating mean squared error
# of prediction vs truth (first line of output)
# and of betahat vs truth (2nd line of output)
[lasso_3, ridge_3]
source='regression_modules/evaluate.R'
input: y = output_from(1)['test'], yhat = output_from(2)['pred'], coef = output_from(2)['coef']
output: f"{_input['yhat']:nn}.mse.csv"
bash: expand = True, stderr = False
Rscript {source} b="c({paths(beta):,})" test="'{_input[0]}'" fpred="'{_input[1]}'" fcoef="'{_input[2]}'" output="'{_output}'"
In [7]:
# Run default core analysis
[default_1]
sos_run(['ridge', 'lasso'])
In [8]:
# Compute and report error estimates
# in HTML table format
[default_2]
depends: executable("pandoc")
input: dynamic("*.mse.csv")
import numpy as np
ridge_summary = np.mean(np.array([sum([x.strip().split() for x in open(f).readlines()], []) for f in _input if 'ridge' in str(f)], dtype = float).T, axis = 1).tolist()
lasso_summary = np.mean(np.array([sum([x.strip().split() for x in open(f).readlines()], []) for f in _input if 'lasso' in str(f)], dtype = float).T, axis = 1).tolist()
report: output = "report.md", expand = "${ }"
%% Comparison summary
| Method | Avg. Estimation Error | Avg. Prediction Error |
|:------:|:-------:|:-------:|
| LASSO | ${lasso_summary[1]} | ${lasso_summary[0]} |
| Ridge | ${ridge_summary[1]} | ${ridge_summary[0]} |
download:
https://vatlab.github.io/sos-docs/css/pandoc.css
pandoc: input = "report.md", output = "report.html", args = '{input:q} --css pandoc.css --self-contained -s --output {output:q}'
In [ ]:
%sosrun
In [ ]:
%preview report.html