В чем разница между «регулярной» линейной регрессией и глубокой обучаемостью?

Я хочу знать разницу между линейной регрессией в обычном анализе машинного обучения и линейной регрессией в условиях «глубокого обучения». Какие алгоритмы используются для линейной регрессии в условиях глубокого обучения.

Предполагая, что под глубоким обучением вы имели в виду более точные нейронные сети: полностью подключенная нейронная сеть с прямой связью и только линейными функциями активации будет выполнять линейную регрессию независимо от количества слоев. Одно из отличий состоит в том, что в нейронной сети обычно используется градиентный спуск, тогда как в случае «нормальной» линейной регрессии используется нормальное уравнение, если это возможно (когда число объектов не слишком велико).

Пример полностью подключенной нейронной сети с прямой связью без скрытого слоя и с использованием функции линейной активации (а именно, функции активации идентификации):

Если вы замените функцию активации выходного слоя сигмовидной функцией, то нейронная сеть выполнит логистическую регрессию. Если вы замените функцию активации выходного слоя функцией softmax и добавите несколько выходных единиц, то нейронная сеть выполнит мультиклассовую логистическую регрессию: разницу между логистической регрессией и нейронными сетями . Если вы замените функцию стоимости на потерю шарнира , то нейронная сеть - это SVM, оптимизированный в своей первичной форме: http://cs231n.github.io/linear-classify/ .

Вот пример, показанный на картинке выше, запрограммированный в TensorFlow:

""" Linear Regression Example """
# https://github.com/tflearn/tflearn/blob/master/examples/basics/linear_regression.py

from __future__ import absolute_import, division, print_function

import tflearn

# Regression data
X = [3.3,4.4,5.5,6.71,6.93,4.168,9.779,6.182,7.59,2.167,7.042,10.791,5.313,7.997,5.654,9.27,3.1]
Y = [1.7,2.76,2.09,3.19,1.694,1.573,3.366,2.596,2.53,1.221,2.827,3.465,1.65,2.904,2.42,2.94,1.3]

# Linear Regression graph
input_ = tflearn.input_data(shape=[None])
linear = tflearn.single_unit(input_)
regression = tflearn.regression(linear, optimizer='sgd', loss='mean_square',
                                metric='R2', learning_rate=0.01)
m = tflearn.DNN(regression)
m.fit(X, Y, n_epoch=1000, show_metric=True, snapshot_epoch=False)

print("\nRegression result:")
print("Y = " + str(m.get_weights(linear.W)) +
      "*X + " + str(m.get_weights(linear.b)))

print("\nTest prediction for x = 3.2, 3.3, 3.4:")
print(m.predict([3.2, 3.3, 3.4]))
# should output (close, not exact) y = [1.5315033197402954, 1.5585315227508545, 1.5855598449707031]

Вот фрагмент кода, который не использует библиотеки нейронной сети:

# From http://briandolhansky.com/blog/artificial-neural-networks-linear-regression-part-1
import matplotlib.pyplot as plt
import numpy as np

# Load the data and create the data matrices X and Y
# This creates a feature vector X with a column of ones (bias)
# and a column of car weights.
# The target vector Y is a column of MPG values for each car.
X_file = np.genfromtxt('mpg.csv', delimiter=',', skip_header=1)
N = np.shape(X_file)[0]
X = np.hstack((np.ones(N).reshape(N, 1), X_file[:, 4].reshape(N, 1)))
Y = X_file[:, 0]

# Standardize the input 
X[:, 1] = (X[:, 1]-np.mean(X[:, 1]))/np.std(X[:, 1])

# There are two weights, the bias weight and the feature weight
w = np.array([0, 0])

# Start batch gradient descent, it will run for max_iter epochs and have a step
# size eta
max_iter = 100
eta = 1E-3
for t in range(0, max_iter):
    # We need to iterate over each data point for one epoch
    grad_t = np.array([0., 0.])
    for i in range(0, N):
        x_i = X[i, :]
        y_i = Y[i]
        # Dot product, computes h(x_i, w)
        h = np.dot(w, x_i)-y_i
        grad_t += 2*x_i*h

    # Update the weights
    w = w - eta*grad_t
print "Weights found:",w

# Plot the data and best fit line
tt = np.linspace(np.min(X[:, 1]), np.max(X[:, 1]), 10)
bf_line = w[0]+w[1]*tt

plt.plot(X[:, 1], Y, 'kx', tt, bf_line, 'r-')
plt.xlabel('Weight (Normalized)')
plt.ylabel('MPG')
plt.title('ANN Regression on 1D MPG Data')

plt.savefig('mpg.png')

plt.show()

Файл данных mpg.csv(~ 50% сокращено из-за ограничения размера ответа в Stack Exchange):

mpg (n),cylinders (n),displacement (n),horsepower (n),weight (n),acceleration (n),year (n),origin (n), name (s)
18.000000,8.000000,307.000000,130.000000,3504.000000,12.000000,70.000000,1.000000
15.000000,8.000000,350.000000,165.000000,3693.000000,11.500000,70.000000,1.000000
18.000000,8.000000,318.000000,150.000000,3436.000000,11.000000,70.000000,1.000000
16.000000,8.000000,304.000000,150.000000,3433.000000,12.000000,70.000000,1.000000
17.000000,8.000000,302.000000,140.000000,3449.000000,10.500000,70.000000,1.000000
15.000000,8.000000,429.000000,198.000000,4341.000000,10.000000,70.000000,1.000000
14.000000,8.000000,454.000000,220.000000,4354.000000,9.000000,70.000000,1.000000
14.000000,8.000000,440.000000,215.000000,4312.000000,8.500000,70.000000,1.000000
14.000000,8.000000,455.000000,225.000000,4425.000000,10.000000,70.000000,1.000000
15.000000,8.000000,390.000000,190.000000,3850.000000,8.500000,70.000000,1.000000
15.000000,8.000000,383.000000,170.000000,3563.000000,10.000000,70.000000,1.000000
14.000000,8.000000,340.000000,160.000000,3609.000000,8.000000,70.000000,1.000000
15.000000,8.000000,400.000000,150.000000,3761.000000,9.500000,70.000000,1.000000
14.000000,8.000000,455.000000,225.000000,3086.000000,10.000000,70.000000,1.000000
24.000000,4.000000,113.000000,95.000000,2372.000000,15.000000,70.000000,3.000000
22.000000,6.000000,198.000000,95.000000,2833.000000,15.500000,70.000000,1.000000
18.000000,6.000000,199.000000,97.000000,2774.000000,15.500000,70.000000,1.000000
21.000000,6.000000,200.000000,85.000000,2587.000000,16.000000,70.000000,1.000000
27.000000,4.000000,97.000000,88.000000,2130.000000,14.500000,70.000000,3.000000
26.000000,4.000000,97.000000,46.000000,1835.000000,20.500000,70.000000,2.000000
25.000000,4.000000,110.000000,87.000000,2672.000000,17.500000,70.000000,2.000000
24.000000,4.000000,107.000000,90.000000,2430.000000,14.500000,70.000000,2.000000
25.000000,4.000000,104.000000,95.000000,2375.000000,17.500000,70.000000,2.000000
26.000000,4.000000,121.000000,113.000000,2234.000000,12.500000,70.000000,2.000000
21.000000,6.000000,199.000000,90.000000,2648.000000,15.000000,70.000000,1.000000
10.000000,8.000000,360.000000,215.000000,4615.000000,14.000000,70.000000,1.000000
10.000000,8.000000,307.000000,200.000000,4376.000000,15.000000,70.000000,1.000000
11.000000,8.000000,318.000000,210.000000,4382.000000,13.500000,70.000000,1.000000
9.000000,8.000000,304.000000,193.000000,4732.000000,18.500000,70.000000,1.000000
27.000000,4.000000,97.000000,88.000000,2130.000000,14.500000,71.000000,3.000000
28.000000,4.000000,140.000000,90.000000,2264.000000,15.500000,71.000000,1.000000
25.000000,4.000000,113.000000,95.000000,2228.000000,14.000000,71.000000,3.000000
19.000000,6.000000,232.000000,100.000000,2634.000000,13.000000,71.000000,1.000000
16.000000,6.000000,225.000000,105.000000,3439.000000,15.500000,71.000000,1.000000
17.000000,6.000000,250.000000,100.000000,3329.000000,15.500000,71.000000,1.000000
19.000000,6.000000,250.000000,88.000000,3302.000000,15.500000,71.000000,1.000000
18.000000,6.000000,232.000000,100.000000,3288.000000,15.500000,71.000000,1.000000
14.000000,8.000000,350.000000,165.000000,4209.000000,12.000000,71.000000,1.000000
14.000000,8.000000,400.000000,175.000000,4464.000000,11.500000,71.000000,1.000000
14.000000,8.000000,351.000000,153.000000,4154.000000,13.500000,71.000000,1.000000
14.000000,8.000000,318.000000,150.000000,4096.000000,13.000000,71.000000,1.000000
12.000000,8.000000,383.000000,180.000000,4955.000000,11.500000,71.000000,1.000000
13.000000,8.000000,400.000000,170.000000,4746.000000,12.000000,71.000000,1.000000
13.000000,8.000000,400.000000,175.000000,5140.000000,12.000000,71.000000,1.000000
18.000000,6.000000,258.000000,110.000000,2962.000000,13.500000,71.000000,1.000000
22.000000,4.000000,140.000000,72.000000,2408.000000,19.000000,71.000000,1.000000
19.000000,6.000000,250.000000,100.000000,3282.000000,15.000000,71.000000,1.000000
18.000000,6.000000,250.000000,88.000000,3139.000000,14.500000,71.000000,1.000000
23.000000,4.000000,122.000000,86.000000,2220.000000,14.000000,71.000000,1.000000
28.000000,4.000000,116.000000,90.000000,2123.000000,14.000000,71.000000,2.000000
30.000000,4.000000,79.000000,70.000000,2074.000000,19.500000,71.000000,2.000000
30.000000,4.000000,88.000000,76.000000,2065.000000,14.500000,71.000000,2.000000
31.000000,4.000000,71.000000,65.000000,1773.000000,19.000000,71.000000,3.000000
35.000000,4.000000,72.000000,69.000000,1613.000000,18.000000,71.000000,3.000000
27.000000,4.000000,97.000000,60.000000,1834.000000,19.000000,71.000000,2.000000
26.000000,4.000000,91.000000,70.000000,1955.000000,20.500000,71.000000,1.000000
24.000000,4.000000,113.000000,95.000000,2278.000000,15.500000,72.000000,3.000000
25.000000,4.000000,97.500000,80.000000,2126.000000,17.000000,72.000000,1.000000
23.000000,4.000000,97.000000,54.000000,2254.000000,23.500000,72.000000,2.000000
20.000000,4.000000,140.000000,90.000000,2408.000000,19.500000,72.000000,1.000000
21.000000,4.000000,122.000000,86.000000,2226.000000,16.500000,72.000000,1.000000
13.000000,8.000000,350.000000,165.000000,4274.000000,12.000000,72.000000,1.000000
14.000000,8.000000,400.000000,175.000000,4385.000000,12.000000,72.000000,1.000000
15.000000,8.000000,318.000000,150.000000,4135.000000,13.500000,72.000000,1.000000
14.000000,8.000000,351.000000,153.000000,4129.000000,13.000000,72.000000,1.000000
17.000000,8.000000,304.000000,150.000000,3672.000000,11.500000,72.000000,1.000000
11.000000,8.000000,429.000000,208.000000,4633.000000,11.000000,72.000000,1.000000
13.000000,8.000000,350.000000,155.000000,4502.000000,13.500000,72.000000,1.000000
12.000000,8.000000,350.000000,160.000000,4456.000000,13.500000,72.000000,1.000000
13.000000,8.000000,400.000000,190.000000,4422.000000,12.500000,72.000000,1.000000
19.000000,3.000000,70.000000,97.000000,2330.000000,13.500000,72.000000,3.000000
15.000000,8.000000,304.000000,150.000000,3892.000000,12.500000,72.000000,1.000000
13.000000,8.000000,307.000000,130.000000,4098.000000,14.000000,72.000000,1.000000
13.000000,8.000000,302.000000,140.000000,4294.000000,16.000000,72.000000,1.000000
14.000000,8.000000,318.000000,150.000000,4077.000000,14.000000,72.000000,1.000000
18.000000,4.000000,121.000000,112.000000,2933.000000,14.500000,72.000000,2.000000
22.000000,4.000000,121.000000,76.000000,2511.000000,18.000000,72.000000,2.000000
21.000000,4.000000,120.000000,87.000000,2979.000000,19.500000,72.000000,2.000000
26.000000,4.000000,96.000000,69.000000,2189.000000,18.000000,72.000000,2.000000
22.000000,4.000000,122.000000,86.000000,2395.000000,16.000000,72.000000,1.000000
28.000000,4.000000,97.000000,92.000000,2288.000000,17.000000,72.000000,3.000000
23.000000,4.000000,120.000000,97.000000,2506.000000,14.500000,72.000000,3.000000
28.000000,4.000000,98.000000,80.000000,2164.000000,15.000000,72.000000,1.000000
27.000000,4.000000,97.000000,88.000000,2100.000000,16.500000,72.000000,3.000000
13.000000,8.000000,350.000000,175.000000,4100.000000,13.000000,73.000000,1.000000
14.000000,8.000000,304.000000,150.000000,3672.000000,11.500000,73.000000,1.000000
13.000000,8.000000,350.000000,145.000000,3988.000000,13.000000,73.000000,1.000000
14.000000,8.000000,302.000000,137.000000,4042.000000,14.500000,73.000000,1.000000
15.000000,8.000000,318.000000,150.000000,3777.000000,12.500000,73.000000,1.000000
12.000000,8.000000,429.000000,198.000000,4952.000000,11.500000,73.000000,1.000000
13.000000,8.000000,400.000000,150.000000,4464.000000,12.000000,73.000000,1.000000
13.000000,8.000000,351.000000,158.000000,4363.000000,13.000000,73.000000,1.000000
14.000000,8.000000,318.000000,150.000000,4237.000000,14.500000,73.000000,1.000000
13.000000,8.000000,440.000000,215.000000,4735.000000,11.000000,73.000000,1.000000
12.000000,8.000000,455.000000,225.000000,4951.000000,11.000000,73.000000,1.000000
13.000000,8.000000,360.000000,175.000000,3821.000000,11.000000,73.000000,1.000000
18.000000,6.000000,225.000000,105.000000,3121.000000,16.500000,73.000000,1.000000
16.000000,6.000000,250.000000,100.000000,3278.000000,18.000000,73.000000,1.000000
18.000000,6.000000,232.000000,100.000000,2945.000000,16.000000,73.000000,1.000000
18.000000,6.000000,250.000000,88.000000,3021.000000,16.500000,73.000000,1.000000
23.000000,6.000000,198.000000,95.000000,2904.000000,16.000000,73.000000,1.000000
26.000000,4.000000,97.000000,46.000000,1950.000000,21.000000,73.000000,2.000000
11.000000,8.000000,400.000000,150.000000,4997.000000,14.000000,73.000000,1.000000
12.000000,8.000000,400.000000,167.000000,4906.000000,12.500000,73.000000,1.000000
13.000000,8.000000,360.000000,170.000000,4654.000000,13.000000,73.000000,1.000000
12.000000,8.000000,350.000000,180.000000,4499.000000,12.500000,73.000000,1.000000
18.000000,6.000000,232.000000,100.000000,2789.000000,15.000000,73.000000,1.000000
20.000000,4.000000,97.000000,88.000000,2279.000000,19.000000,73.000000,3.000000
21.000000,4.000000,140.000000,72.000000,2401.000000,19.500000,73.000000,1.000000
22.000000,4.000000,108.000000,94.000000,2379.000000,16.500000,73.000000,3.000000
18.000000,3.000000,70.000000,90.000000,2124.000000,13.500000,73.000000,3.000000
19.000000,4.000000,122.000000,85.000000,2310.000000,18.500000,73.000000,1.000000
21.000000,6.000000,155.000000,107.000000,2472.000000,14.000000,73.000000,1.000000
26.000000,4.000000,98.000000,90.000000,2265.000000,15.500000,73.000000,2.000000
15.000000,8.000000,350.000000,145.000000,4082.000000,13.000000,73.000000,1.000000
16.000000,8.000000,400.000000,230.000000,4278.000000,9.500000,73.000000,1.000000
29.000000,4.000000,68.000000,49.000000,1867.000000,19.500000,73.000000,2.000000
24.000000,4.000000,116.000000,75.000000,2158.000000,15.500000,73.000000,2.000000
20.000000,4.000000,114.000000,91.000000,2582.000000,14.000000,73.000000,2.000000
19.000000,4.000000,121.000000,112.000000,2868.000000,15.500000,73.000000,2.000000
15.000000,8.000000,318.000000,150.000000,3399.000000,11.000000,73.000000,1.000000
24.000000,4.000000,121.000000,110.000000,2660.000000,14.000000,73.000000,2.000000
20.000000,6.000000,156.000000,122.000000,2807.000000,13.500000,73.000000,3.000000
11.000000,8.000000,350.000000,180.000000,3664.000000,11.000000,73.000000,1.000000
20.000000,6.000000,198.000000,95.000000,3102.000000,16.500000,74.000000,1.000000
19.000000,6.000000,232.000000,100.000000,2901.000000,16.000000,74.000000,1.000000
15.000000,6.000000,250.000000,100.000000,3336.000000,17.000000,74.000000,1.000000
31.000000,4.000000,79.000000,67.000000,1950.000000,19.000000,74.000000,3.000000
26.000000,4.000000,122.000000,80.000000,2451.000000,16.500000,74.000000,1.000000
32.000000,4.000000,71.000000,65.000000,1836.000000,21.000000,74.000000,3.000000
25.000000,4.000000,140.000000,75.000000,2542.000000,17.000000,74.000000,1.000000
16.000000,6.000000,250.000000,100.000000,3781.000000,17.000000,74.000000,1.000000
16.000000,6.000000,258.000000,110.000000,3632.000000,18.000000,74.000000,1.000000
18.000000,6.000000,225.000000,105.000000,3613.000000,16.500000,74.000000,1.000000
16.000000,8.000000,302.000000,140.000000,4141.000000,14.000000,74.000000,1.000000
13.000000,8.000000,350.000000,150.000000,4699.000000,14.500000,74.000000,1.000000
14.000000,8.000000,318.000000,150.000000,4457.000000,13.500000,74.000000,1.000000
14.000000,8.000000,302.000000,140.000000,4638.000000,16.000000,74.000000,1.000000
14.000000,8.000000,304.000000,150.000000,4257.000000,15.500000,74.000000,1.000000
29.000000,4.000000,98.000000,83.000000,2219.000000,16.500000,74.000000,2.000000
26.000000,4.000000,79.000000,67.000000,1963.000000,15.500000,74.000000,2.000000
26.000000,4.000000,97.000000,78.000000,2300.000000,14.500000,74.000000,2.000000
31.000000,4.000000,76.000000,52.000000,1649.000000,16.500000,74.000000,3.000000
32.000000,4.000000,83.000000,61.000000,2003.000000,19.000000,74.000000,3.000000
28.000000,4.000000,90.000000,75.000000,2125.000000,14.500000,74.000000,1.000000
24.000000,4.000000,90.000000,75.000000,2108.000000,15.500000,74.000000,2.000000
26.000000,4.000000,116.000000,75.000000,2246.000000,14.000000,74.000000,2.000000
24.000000,4.000000,120.000000,97.000000,2489.000000,15.000000,74.000000,3.000000
26.000000,4.000000,108.000000,93.000000,2391.000000,15.500000,74.000000,3.000000
31.000000,4.000000,79.000000,67.000000,2000.000000,16.000000,74.000000,2.000000
19.000000,6.000000,225.000000,95.000000,3264.000000,16.000000,75.000000,1.000000
18.000000,6.000000,250.000000,105.000000,3459.000000,16.000000,75.000000,1.000000
15.000000,6.000000,250.000000,72.000000,3432.000000,21.000000,75.000000,1.000000
15.000000,6.000000,250.000000,72.000000,3158.000000,19.500000,75.000000,1.000000
16.000000,8.000000,400.000000,170.000000,4668.000000,11.500000,75.000000,1.000000
15.000000,8.000000,350.000000,145.000000,4440.000000,14.000000,75.000000,1.000000
16.000000,8.000000,318.000000,150.000000,4498.000000,14.500000,75.000000,1.000000
14.000000,8.000000,351.000000,148.000000,4657.000000,13.500000,75.000000,1.000000
17.000000,6.000000,231.000000,110.000000,3907.000000,21.000000,75.000000,1.000000
16.000000,6.000000,250.000000,105.000000,3897.000000,18.500000,75.000000,1.000000
15.000000,6.000000,258.000000,110.000000,3730.000000,19.000000,75.000000,1.000000
18.000000,6.000000,225.000000,95.000000,3785.000000,19.000000,75.000000,1.000000
21.000000,6.000000,231.000000,110.000000,3039.000000,15.000000,75.000000,1.000000
20.000000,8.000000,262.000000,110.000000,3221.000000,13.500000,75.000000,1.000000
13.000000,8.000000,302.000000,129.000000,3169.000000,12.000000,75.000000,1.000000
29.000000,4.000000,97.000000,75.000000,2171.000000,16.000000,75.000000,3.000000
23.000000,4.000000,140.000000,83.000000,2639.000000,17.000000,75.000000,1.000000
20.000000,6.000000,232.000000,100.000000,2914.000000,16.000000,75.000000,1.000000
23.000000,4.000000,140.000000,78.000000,2592.000000,18.500000,75.000000,1.000000
24.000000,4.000000,134.000000,96.000000,2702.000000,13.500000,75.000000,3.000000
25.000000,4.000000,90.000000,71.000000,2223.000000,16.500000,75.000000,2.000000
24.000000,4.000000,119.000000,97.000000,2545.000000,17.000000,75.000000,3.000000
18.000000,6.000000,171.000000,97.000000,2984.000000,14.500000,75.000000,1.000000
29.000000,4.000000,90.000000,70.000000,1937.000000,14.000000,75.000000,2.000000
19.000000,6.000000,232.000000,90.000000,3211.000000,17.000000,75.000000,1.000000
23.000000,4.000000,115.000000,95.000000,2694.000000,15.000000,75.000000,2.000000
23.000000,4.000000,120.000000,88.000000,2957.000000,17.000000,75.000000,2.000000
22.000000,4.000000,121.000000,98.000000,2945.000000,14.500000,75.000000,2.000000
25.000000,4.000000,121.000000,115.000000,2671.000000,13.500000,75.000000,2.000000
33.000000,4.000000,91.000000,53.000000,1795.000000,17.500000,75.000000,3.000000
28.000000,4.000000,107.000000,86.000000,2464.000000,15.500000,76.000000,2.000000
25.000000,4.000000,116.000000,81.000000,2220.000000,16.900000,76.000000,2.000000
25.000000,4.000000,140.000000,92.000000,2572.000000,14.900000,76.000000,1.000000
26.000000,4.000000,98.000000,79.000000,2255.000000,17.700000,76.000000,1.000000
27.000000,4.000000,101.000000,83.000000,2202.000000,15.300000,76.000000,2.000000
17.500000,8.000000,305.000000,140.000000,4215.000000,13.000000,76.000000,1.000000
16.000000,8.000000,318.000000,150.000000,4190.000000,13.000000,76.000000,1.000000
15.500000,8.000000,304.000000,120.000000,3962.000000,13.900000,76.000000,1.000000
14.500000,8.000000,351.000000,152.000000,4215.000000,12.800000,76.000000,1.000000
22.000000,6.000000,225.000000,100.000000,3233.000000,15.400000,76.000000,1.000000
22.000000,6.000000,250.000000,105.000000,3353.000000,14.500000,76.000000,1.000000
24.000000,6.000000,200.000000,81.000000,3012.000000,17.600000,76.000000,1.000000
22.500000,6.000000,232.000000,90.000000,3085.000000,17.600000,76.000000,1.000000
29.000000,4.000000,85.000000,52.000000,2035.000000,22.200000,76.000000,1.000000
24.500000,4.000000,98.000000,60.000000,2164.000000,22.100000,76.000000,1.000000
29.000000,4.000000,90.000000,70.000000,1937.000000,14.200000,76.000000,2.000000
33.000000,4.000000,91.000000,53.000000,1795.000000,17.400000,76.000000,3.000000
20.000000,6.000000,225.000000,100.000000,3651.000000,17.700000,76.000000,1.000000
18.000000,6.000000,250.000000,78.000000,3574.000000,21.000000,76.000000,1.000000
18.500000,6.000000,250.000000,110.000000,3645.000000,16.200000,76.000000,1.000000
17.500000,6.000000,258.000000,95.000000,3193.000000,17.800000,76.000000,1.000000
29.500000,4.000000,97.000000,71.000000,1825.000000,12.200000,76.000000,2.000000
32.000000,4.000000,85.000000,70.000000,1990.000000,17.000000,76.000000,3.000000
28.000000,4.000000,97.000000,75.000000,2155.000000,16.400000,76.000000,3.000000
26.500000,4.000000,140.000000,72.000000,2565.000000,13.600000,76.000000,1.000000
20.000000,4.000000,130.000000,102.000000,3150.000000,15.700000,76.000000,2.000000
13.000000,8.000000,318.000000,150.000000,3940.000000,13.200000,76.000000,1.000000
19.000000,4.000000,120.000000,88.000000,3270.000000,21.900000,76.000000,2.000000
19.000000,6.000000,156.000000,108.000000,2930.000000,15.500000,76.000000,3.000000
16.500000,6.000000,168.000000,120.000000,3820.000000,16.700000,76.000000,2.000000
16.500000,8.000000,350.000000,180.000000,4380.000000,12.100000,76.000000,1.000000
13.000000,8.000000,350.000000,145.000000,4055.000000,12.000000,76.000000,1.000000
13.000000,8.000000,302.000000,130.000000,3870.000000,15.000000,76.000000,1.000000
13.000000,8.000000,318.000000,150.000000,3755.000000,14.000000,76.000000,1.000000
31.500000,4.000000,98.000000,68.000000,2045.000000,18.500000,77.000000,3.000000
30.000000,4.000000,111.000000,80.000000,2155.000000,14.800000,77.000000,1.000000
36.000000,4.000000,79.000000,58.000000,1825.000000,18.600000,77.000000,2.000000
25.500000,4.000000,122.000000,96.000000,2300.000000,15.500000,77.000000,1.000000
33.500000,4.000000,85.000000,70.000000,1945.000000,16.800000,77.000000,3.000000
17.500000,8.000000,305.000000,145.000000,3880.000000,12.500000,77.000000,1.000000
17.000000,8.000000,260.000000,110.000000,4060.000000,19.000000,77.000000,1.000000
15.500000,8.000000,318.000000,145.000000,4140.000000,13.700000,77.000000,1.000000
15.000000,8.000000,302.000000,130.000000,4295.000000,14.900000,77.000000,1.000000
17.500000,6.000000,250.000000,110.000000,3520.000000,16.400000,77.000000,1.000000
20.500000,6.000000,231.000000,105.000000,3425.000000,16.900000,77.000000,1.000000
19.000000,6.000000,225.000000,100.000000,3630.000000,17.700000,77.000000,1.000000
18.500000,6.000000,250.000000,98.000000,3525.000000,19.000000,77.000000,1.000000
16.000000,8.000000,400.000000,180.000000,4220.000000,11.100000,77.000000,1.000000
15.500000,8.000000,350.000000,170.000000,4165.000000,11.400000,77.000000,1.000000
15.500000,8.000000,400.000000,190.000000,4325.000000,12.200000,77.000000,1.000000
16.000000,8.000000,351.000000,149.000000,4335.000000,14.500000,77.000000,1.000000
29.000000,4.000000,97.000000,78.000000,1940.000000,14.500000,77.000000,2.000000
24.500000,4.000000,151.000000,88.000000,2740.000000,16.000000,77.000000,1.000000
26.000000,4.000000,97.000000,75.000000,2265.000000,18.200000,77.000000,3.000000
25.500000,4.000000,140.000000,89.000000,2755.000000,15.800000,77.000000,1.000000
30.500000,4.000000,98.000000,63.000000,2051.000000,17.000000,77.000000,1.000000
33.500000,4.000000,98.000000,83.000000,2075.000000,15.900000,77.000000,1.000000
30.000000,4.000000,97.000000,67.000000,1985.000000,16.400000,77.000000,3.000000
30.500000,4.000000,97.000000,78.000000,2190.000000,14.100000,77.000000,2.000000
22.000000,6.000000,146.000000,97.000000,2815.000000,14.500000,77.000000,3.000000
21.500000,4.000000,121.000000,110.000000,2600.000000,12.800000,77.000000,2.000000
21.500000,3.000000,80.000000,110.000000,2720.000000,13.500000,77.000000,3.000000
43.100000,4.000000,90.000000,48.000000,1985.000000,21.500000,78.000000,2.000000
36.100000,4.000000,98.000000,66.000000,1800.000000,14.400000,78.000000,1.000000
32.800000,4.000000,78.000000,52.000000,1985.000000,19.400000,78.000000,3.000000
39.400000,4.000000,85.000000,70.000000,2070.000000,18.600000,78.000000,3.000000
36.100000,4.000000,91.000000,60.000000,1800.000000,16.400000,78.000000,3.000000
19.900000,8.000000,260.000000,110.000000,3365.000000,15.500000,78.000000,1.000000

Для регрессии, которая для глубокого обучения является нелинейной в большинстве случаев, конечный уровень имеет 1 нейрон с функцией тождественности и функцией потерь, которую мы оптимизируем, это MSE, MAE вместо двоичной или категориальной кросс-энтропии, используемой для классификации.