X = X.reshape(-1, 1) y = y.reshape(-1, 1) # creating model model = LinearRegression() # Fitting training data model.fit(X, y) # Printing theta0 and theta1 # intercept_ --> theta0 and coef_ --> list of [theta1, theta2, ...] print(model.intercept_, model.coef_) # Coeff of determination for training data print(model.score(X, y)) # predicting for x = 3.5 and x = 7.0 print(model.predict(np.array([3.5, 7.0]).reshape(-1, 1))) # Multi-feature training set fname = join(HOME, path, './multi-feature.txt') X1, X2, y = np.loadtxt(fname, delimiter=',', unpack=True) # X = np.hstack((X1, X2)) X = np.c_[X1, X2] X = X.reshape(-1, 2) y = y.reshape(-1, 1) model.normalize = True model.fit(X, y) print(model.intercept_, model.coef_, model.score(X, y)) print(model.predict(np.array([1650, 3]).reshape(-1, 2))) # Using Polynomial features X_ = PolynomialFeatures(degree=2, include_bias=True).fit_transform(X) model.fit(X_, y) print(model.intercept_, model.coef_, model.score(X_, y))
def linearRegression(dataframe_with_efficacy,
efficacy=True,
normalize=False,
include_position=False,
testsize=0.25):
df = dataframe_with_efficacy.copy()
df = df.drop(columns=['rep', 'field_id'])
if include_position:
pass
else:
df = df.drop(columns=['row', 'range'])
df['sym'] = df['sym'].astype('category')
df = pd.get_dummies(df)
df = df.loc[df['sym_ctrl'] == False]
df = df.drop(columns='sym_ctrl')
if efficacy:
target = 'efficacy_in_percent'
df = df.drop(columns=['value'])
else:
target = 'value'
df = df.drop(columns=['efficacy_in_percent'])
features = df.drop(columns=target).columns
# Make training set. Leave out 25% data for testing.
# create multiple linear regression object
mlr = LinearRegression(fit_intercept=True)
# Whether or not to normalize:
mlr.normalize = normalize
# Separate into 75% train and 25% test:
# Test size is default 25%
x_train, x_test, y_train, y_test = train_test_split(df[features],
df[target],
test_size=testsize,
shuffle=True)
# fit linear regression
mlr.fit(x_train, y_train)
# get the slope and intercept of the line best fit.
# print(mlr.intercept_)
print('features in order of decreasing value of coeficients:')
print('feature: coefficient value; target: ', target)
print('--------------------------')
sorted_idx = np.argsort(mlr.coef_)
f = features[sorted_idx[::-1]]
c = mlr.coef_[sorted_idx[::-1]]
for cidx, ff in enumerate(f):
print(ff, ": ", c[cidx])
# Run the model on the test set, plot the comparison.
y_prediction = mlr.predict(x_test)
rmse_model = (np.mean(y_prediction - y_test)**2)**0.5
cv4 = cross_val_score(mlr,
df[features],
df[target],
cv=4,
scoring="neg_mean_squared_error")
rmse_model_cross_fold_4 = (np.mean(np.sign(cv4) * cv4))**0.5
print('')
print('')
print('--------------------------------------')
print('rmse_for_cross_val_four_times: ',
np.round(rmse_model_cross_fold_4, 2))
print('--------------------------------------')
print('')
print('')
print('')
print('Comparing predicted vs truth value in dataset')
print('---------------------------------------------------------')
fig = plt.figure(figsize=(6, 6))
plt.scatter(y_test, y_prediction)
plt.title('predicted vs test: RMSE = %s [units]' %
(np.round(rmse_model, 2)))
plt.ylabel('predicted value for %s' % (target))
plt.xlabel('test value for %s' % (target))
return mlr
