caravan_regression.py

import sys
import os.path
import caravan_dump

import numpy as np
from sklearn.linear_model import Ridge
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import PolynomialFeatures
from sklearn.model_selection import train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.externals import joblib

if len(sys.argv) != 2:
    sys.stderr.write("Invalid argument\n")
    sys.stderr.write("  Usage: python %s dump.bin\n" % os.path.basename(__file__))
    raise Exception("Invalid command line argument")

dump = caravan_dump.CaravanDump( sys.argv[1] )

def parse_results(dump):
    x = []
    y = []
    for r in dump.runs:
        result = r["result"]
        if result[0] == 0.0:
            sys.stderr.write("skipped\n")
            continue
        y.append( result )
        psid = r["parentPSId"]
        ps = dump.parameter_sets[psid]
        x.append( [float(x) for x in ps["point"]] )

    return (np.array(x), np.array(y) )


X,ys = parse_results( dump )
sys.stderr.write("parsed\n")

for k in range(ys.shape[1]):
    y = ys[:,k]

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.4, random_state=1234)

    pipe = make_pipeline( StandardScaler(), PolynomialFeatures(degree=5), Ridge(alpha=0.005, fit_intercept=True))

    from sklearn.model_selection import GridSearchCV

    params = dict(polynomialfeatures__degree=[3,4,5], ridge__alpha= np.logspace(-5,2,num=8) )
    grid_search = GridSearchCV( pipe, param_grid=params, cv=5, n_jobs=4 )
    grid_search.fit( X_train, y_train )

    print( grid_search.best_estimator_ )
    print( grid_search.score( X_test, y_test ) )
    joblib.dump( grid_search, "reg_model%d.pkl"%k )