AI-Solar/linear_models.py at master · gokart23/AI-Solar · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
import data
import config
import metrics
import numpy as np
import matplotlib
import sklearn.svm as svm
import matplotlib.pyplot as plt
from sklearn import linear_model
import sklearn.ensemble as ensemble
from sklearn.model_selection import GridSearchCV


def baseline_model(city='', savefile="models.baseline_model"):
    """Train a linear regression model without regularization and obtain predictions on test-set"""

    if not city:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset()
    else:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset(config.datasets[city])

    regr_lm = linear_model.LinearRegression()

    if city:
        print('\nFitting model for ' + city)
    else:
        print("\nFitting global model")
    regr_lm.fit(X_train, y_train)
    # print('Coefficients shape: \n', regr_lm.coef_.shape)

    print("Predicting values")
    pred_val = regr_lm.predict(X_val)
    pred_test = regr_lm.predict(X_test)
    results_heading = 'Baseline Model - Linear Regression without regularization'
    return metrics.generate_results(city, y_val, y_test, pred_val, pred_test, savefile, results_heading)


def lasso_model(city='', max_degree=4, savefile="models.lasso_model"):
    """Train a linear regression model with L1-regularization (and polynomial kernel),
       validate according to alpha, and obtain predictions on test-set"""

    if not city:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset()
    else:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset(config.datasets[city])

    for cur_degree in range(2, max_degree):
        for dataset in [X_train, X_val, X_test]:
            dataset = np.hstack( (dataset, np.power(dataset, cur_degree) ) )

    if city:
        print('\nFitting model for ' + city)
    else:
        print("\nFitting global model")


    max_alpha, max_r2 = 0, 0
    for alpha in [0.00001, 0.0001, 0.0005, 0.001, 0.01, 0.1]:
        regr_lm = linear_model.Lasso(alpha=alpha, max_iter=1e6)

        regr_lm.fit(X_train, y_train)
        print("Predicting values on", alpha)
        pred_val = regr_lm.predict(X_val)
        pred_test = regr_lm.predict(X_test)

        results_heading = 'Linear Regression with L1-regularization'
        results_heading += "\nCurrent alpha:" + str(alpha)

        rmse_val, r2_val, _, _ = metrics.generate_results(city, y_val, y_test, pred_val, pred_test, savefile + "_" + str(alpha), results_heading)
        max_alpha, max_r2 = (alpha if max_r2 < r2_val else max_alpha), (r2_val if max_r2 < r2_val else max_r2)

    # TODO: Change this to just rename instead of recompute
    print ("Best model observed on validation for alpha", max_alpha, "with score", max_r2)
    regr_lm = linear_model.Lasso(alpha=max_alpha, max_iter=1e8)

    regr_lm.fit(X_train, y_train)
    print("Predicting values on best model")
    pred_val = regr_lm.predict(X_val)
    pred_test = regr_lm.predict(X_test)

    results_heading = 'Baseline Model - Linear Regression with L1-regularization'
    results_heading += "\nCurrent alpha:" + str(alpha)

    metrics.generate_results(city, y_val, y_test, pred_val, pred_test, savefile + "_best", results_heading)

    plt.figure()
    plt.plot(regr_lm.coef_[0, :], color='gold', linewidth=2, label='Lasso coefficients')
    plt.savefig(savefile + "_best_coef_plot.png")
    plt.close()

    return max_alpha, max_r2

def svr_model(city='', max_degree=4, savefile="models.svr_model"):
    """Train a support vector regression (SVR) model with different kernels,
       validate according to alpha, and obtain predictions on test-set"""

    if not city:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset()
    else:
        X_train, y_train, X_val, y_val, X_test, y_test = data.load_dataset(config.datasets[city])

    if city:
        print('\nFitting model for ' + city)
    else:
        print("\nFitting global model")

    for predictor_idx in range( y_train.shape[1] ):
        best_model, best_C, best_r2 = 0, 0, 0
        for C in [0.1, 1, 100, 1000]:
            max_model, max_r2 = 0, 0
            svr_rbf = svm.SVR(kernel='rbf', C=C)
            svr_lin = svm.SVR(kernel='linear', C=C)
            svr_poly = svm.SVR(kernel='poly', C=C)

            for idx, regr_lm in enumerate([svr_rbf, svr_lin, svr_poly]):
                regr_lm.fit(X_train, y_train.ix[:, predictor_idx])
                print("Predicting values on", C)
                pred_val = regr_lm.predict(X_val)
                pred_test = regr_lm.predict(X_test)

                results_heading = 'Linear Regression with L1-regularization'
                results_heading += "\nCurrent predictor:" + str(predictor_idx)

                rmse_val, r2_val, _, _ = metrics.generate_results(city, y_val.ix[:, predictor_idx], y_test.ix[:, predictor_idx], pred_val, pred_test, savefile + "_" + str(C) + "_" + str(predictor_idx), results_heading, skip_save=True)

                print()
                max_model, max_r2 = (idx if max_r2 < r2_val else max_model), (r2_val if max_r2 < r2_val else max_r2)

            best_C, best_model, best_r2 = (C if max_r2 > best_r2 else best_C), (max_model if  max_r2 > best_r2 else best_model), (r2_val if  max_r2 > best_r2 else best_r2)

        print ("Best model observed on validation for C", best_C, "with score", best_r2, "is ", best_model)

        regr_lm = list([ svm.SVR(kernel='rbf', C=best_C), svm.SVR(kernel='linear', C=best_C), svm.SVR(kernel='poly', C=best_C) ])[ best_model ]

        regr_lm.fit(X_train, y_train.ix[:, predictor_idx])
        print("Predicting values on best model")
        pred_val = regr_lm.predict(X_val)
        pred_test = regr_lm.predict(X_test)

        results_heading = 'Support Vector Regression'
        results_heading += "\nCurrent C, model and predictor_idx:" + str(best_C) + "," + str(best_model) + "," + str(predictor_idx)

        metrics.generate_results(city, y_val.ix[:, predictor_idx], y_test.ix[:, predictor_idx], pred_val, pred_test, savefile + "_best_" + str(predictor_idx), results_heading)

# if __name__ == "__main__":
    # metrics.get_model_results(baseline_model)

    # metrics.get_model_results(lasso_model)