vosbs

test = df[17500:]
X_test = test[['speed', 'thickness', 'delta']]
y_test = test[['I']]
X_test_minmax = mm_scaler.transform(X_test[['thickness', 'delta']])
kn = neigh.kneighbors(X_test_minmax, return_distance=False)
test = test.assign(ml_speed=[X_train_clean.loc[i, 'speed'].quantile(0.8) for i in kn])
test = test.assign(I_new_predict=bst.predict(test[['ml_speed', 'thickness', 'delta']]))
test.loc[:, 'ml_speed'] = test['ml_speed'] + (limit - 600 * 3 - test['I_new_predict']) / step
test.loc[:, 'I_new'] = test['I'] + (test['ml_speed'] - test['speed']) * 2000
print((test.I_new > limit).sum() / test.shape[0])

vosbs

dev.loc[:, 'ml_speed'] = dev['ml_speed'] + (limit - 600 * 3 - dev['I_new_predict']) / step
dev.loc[:, 'I_new'] = dev['I'] + (dev['ml_speed'] - dev['speed']) * 2000
print((dev.I_new > limit).sum() / dev.shape[0])
print(dev.ml_speed.mean() / dev.speed.mean())

vosbs

coefs = {}
for i in range(6, 19):
    train_new = train.copy()
    train_new['bin'] = ((resc.predict(Xc) < (i + 2)) & (resc.predict(Xc) >= i)).astype(int)
    Xcb = train_new[train_new.bin == 1][['speed', 'thickness', 'delta']]
    Xcb = sm.add_constant(Xcb)
    Ycb = train_new[train_new.bin == 1][['I']]
    modcb = sm.OLS(Ycb, Xcb)
    rescb = modcb.fit()
    coefs[i] = rescb.params['speed']

vosbs

Xc = train[['thickness', 'delta']]
Xc = sm.add_constant(Xc)
Yc = train[['speed']]
modc = sm.OLS(Yc, Xc)
resc = modc.fit()

vosbs

X_dev_minmax = mm_scaler.transform(X_dev[['thickness', 'delta']])
kn = neigh.kneighbors(X_dev_minmax, return_distance=False)
dev = dev.assign(ml_speed=[X_train_clean.loc[i, 'speed'].quantile(0.8) for i in kn])
print(dev.ml_speed.mean() / dev.speed.mean())
dev = dev.assign(I_new=dev['I'] + (dev['ml_speed'] - dev['speed']) * 2000)
dev = dev.assign(I_new_predict=bst.predict(dev[['ml_speed', 'thickness', 'delta']]))
print((dev.loc[:, 'I_new'] > limit).sum() / dev.shape[0])

vosbs

mm_scaler = preprocessing.MinMaxScaler()
X_train_minmax = mm_scaler.fit_transform(X_train[['thickness', 'delta']])
neigh = KNeighborsRegressor(n_neighbors=20)
X_train_clean = X_train[y_train['I'] < limit].reset_index(drop=True)
X_train_clean_minmax = mm_scaler.transform(X_train_clean[['thickness', 'delta']])
neigh.fit(X_train_clean_minmax, X_train_clean[['speed']])

vosbs

test = test.assign(I_new=test['I'] + (test['ml_speed'] - test['speed']) * 2000)
print((test.I_new > limit).sum() / test.shape[0])

vosbs

def speed_choose(thickness, delta, model=bst, prelimit=limit - 600 * 3):
    speeds = list(range(4, 22))
    speeds.reverse()
    for speed in speeds:
        df = pd.DataFrame({'speed': [speed], 'thickness': [thickness], 'delta': [delta]})
        if model.predict(df) < prelimit:
            return speed
    return 4

vosbs

df['I'] = 3000 + 1500 * (df['thickness'] ** 0.8 + 5 * df['delta'] ** 1.4) + 2000 * df['speed']

vosbs

df['speed'] = 25 - 0.1 * df['thickness'] ** 0.9 - 4 * df['delta'] ** 1.1