PyTorch 多层感知机，数据集来自：nivedithabhandary/HR-Analytics

数据预处理

1
2
3
4
5
6
7
8
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
from sklearn.model_selection import train_test_split
from torch.utils.data import TensorDataset

torch.manual_seed(33)
<torch._C.Generator at 0x7fabe84e8e90>

1
path = "/workspace/disk1/datasets/scalar/HR_comma_sep.csv"
1
2
data = pd.read_csv(path)
data.head(3)

satisfaction_level last_evaluation number_project average_montly_hours time_spend_company Work_accident left promotion_last_5years sales salary
0 0.38 0.53 2 157 3 0 1 0 sales low
1 0.80 0.86 5 262 6 0 1 0 sales medium
2 0.11 0.88 7 272 4 0 1 0 sales medium 
1
2
data = data.rename(columns={"sales": "part"})
# data.head(3)
1
data.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 14999 entries, 0 to 14998
Data columns (total 10 columns):
 #   Column                 Non-Null Count  Dtype  
---  ------                 --------------  -----  
 0   satisfaction_level     14999 non-null  float64
 1   last_evaluation        14999 non-null  float64
 2   number_project         14999 non-null  int64  
 3   average_montly_hours   14999 non-null  int64  
 4   time_spend_company     14999 non-null  int64  
 5   Work_accident          14999 non-null  int64  
 6   left                   14999 non-null  int64  
 7   promotion_last_5years  14999 non-null  int64  
 8   part                   14999 non-null  object 
 9   salary                 14999 non-null  object 
dtypes: float64(2), int64(6), object(2)
memory usage: 1.1+ MB

1
data.part.unique()
array(['sales', 'accounting', 'hr', 'technical', 'support', 'management',
       'IT', 'product_mng', 'marketing', 'RandD'], dtype=object)

1
data.salary.unique()
array(['low', 'medium', 'high'], dtype=object)

1
data.groupby(["salary", "part"]).size()
salary  part       
high    IT               83
        RandD            51
        accounting       74
        hr               45
        management      225
        marketing        80
        product_mng      68
        sales           269
        support         141
        technical       201
low     IT              609
        RandD           364
        accounting      358
        hr              335
        management      180
        marketing       402
        product_mng     451
        sales          2099
        support        1146
        technical      1372
medium  IT              535
        RandD           372
        accounting      335
        hr              359
        management      225
        marketing       376
        product_mng     383
        sales          1772
        support         942
        technical      1147
dtype: int64

1
2
3
4
data = data.join(pd.get_dummies(data.salary))
del data["salary"]
data = data.join(pd.get_dummies(data.part))
del data["part"]
1
data.head(3)

satisfaction_level last_evaluation number_project average_montly_hours time_spend_company Work_accident left promotion_last_5years high low ... IT RandD accounting hr management marketing product_mng sales support technical
0 0.38 0.53 2 157 3 0 1 0 0 1 ... 0 0 0 0 0 0 0 1 0 0
1 0.80 0.86 5 262 6 0 1 0 0 0 ... 0 0 0 0 0 0 0 1 0 0
2 0.11 0.88 7 272 4 0 1 0 0 0 ... 0 0 0 0 0 0 0 1 0 0

3 rows × 21 columns

1
data.columns
Index(['satisfaction_level', 'last_evaluation', 'number_project',
       'average_montly_hours', 'time_spend_company', 'Work_accident', 'left',
       'promotion_last_5years', 'high', 'low', 'medium', 'IT', 'RandD',
       'accounting', 'hr', 'management', 'marketing', 'product_mng', 'sales',
       'support', 'technical'],
      dtype='object')

1
data.shape
(14999, 21)

1
data.left.unique()
array([1, 0])

1
2
# 数据不均衡
data.left.value_counts()
0    11428
1     3571
Name: left, dtype: int64

1
data.left.value_counts() / len(data)
0    0.761917
1    0.238083
Name: left, dtype: float64

1
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
1
2
Y_data = data.left.values.reshape(-1, 1)
X_data = data[[c for c in data.columns if c != "left"]].values
1
2
3
4
5
6
7
8
9
# 训练集和验证集划分
X_train, X_test, Y_train, Y_test = train_test_split(
    X_data, Y_data, train_size=0.8, random_state=33
)

X_train_ = torch.from_numpy(X_train).type(torch.float).to(device)
Y_train_ = torch.from_numpy(Y_train).type(torch.float).to(device)
X_test_ = torch.from_numpy(X_test).type(torch.float).to(device)
Y_test_ = torch.from_numpy(Y_test).type(torch.float).to(device)
1
HRdataset = TensorDataset(X_train_, Y_train_)

创建模型

1
2
3
4
import torch.nn.functional as F
import visdom
from torch import nn, optim
from torch.utils.data import DataLoader
1
2
3
4
5
6
7
8
9
10
11
12
class MLPModel(nn.Module):
    def __init__(self):
        super(MLPModel, self).__init__()
        self.linear_1 = nn.Linear(20, 64)
        self.linear_2 = nn.Linear(64, 64)
        self.linear_3 = nn.Linear(64, 1)

    def forward(self, inputs):
        outputs = F.relu(self.linear_1(inputs))
        outputs = F.relu(self.linear_2(outputs))
        outputs = torch.sigmoid(self.linear_3(outputs))
        return outputs
1
2
3
model = MLPModel()
model.to(device)
model
MLPModel(
  (linear_1): Linear(in_features=20, out_features=64, bias=True)
  (linear_2): Linear(in_features=64, out_features=64, bias=True)
  (linear_3): Linear(in_features=64, out_features=1, bias=True)
)

1
optimizer = optim.Adam(model.parameters(), lr=0.0001)
1
loss_fn = nn.BCELoss()
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
# 使用 pytorch 可视化模块 visdom 可视化训练损失变化情况
viz = visdom.Visdom(
    server="http://localhost",
    port=8097,
    base_url="/visdom",
    username="jinzhongxu",
    password="123123",
)
win = "Multilayer Perceptron loss"
opts = dict(
    title="train_losses",
    xlabel="epoch",
    ylabel="loss",
    markers=True,
    legend=[
        "loss",
    ],
)
viz.line(
    [
        [
            0.0,
        ]
    ],
    [0.0],
    win=win,
    opts=opts,
)
Setting up a new session...





'Multilayer Perceptron loss'

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
batch = 64
epochs = 1000

HRdataloader = DataLoader(HRdataset, batch_size=batch, shuffle=True)

for epoch in range(epochs):
    for x, y in HRdataloader:
        y_pred = model(x)
        loss = loss_fn(y_pred, y)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
    if epoch % 10 == 0:
        with torch.no_grad():
            val_loss = loss_fn(model(X_test_), Y_test_).data.item()
            print(f"epoch: {epoch}, loss:{val_loss}")
            viz.line(
                [
                    [
                        val_loss,
                    ]
                ],
                [epoch],
                win=win,
                update="append",
            )
epoch: 0, loss:0.5702827572822571
epoch: 10, loss:0.5380138754844666
epoch: 20, loss:0.48325881361961365
epoch: 30, loss:0.43280646204948425
epoch: 40, loss:0.40077832341194153
epoch: 50, loss:0.3730575442314148
epoch: 60, loss:0.3515182137489319
epoch: 70, loss:0.333321213722229
epoch: 80, loss:0.3213912844657898
epoch: 90, loss:0.3078565299510956
epoch: 100, loss:0.3066919445991516
epoch: 110, loss:0.28822797536849976
epoch: 120, loss:0.2822719216346741
epoch: 130, loss:0.2698499262332916
epoch: 140, loss:0.2673708200454712
epoch: 150, loss:0.2626533508300781
epoch: 160, loss:0.25525325536727905
epoch: 170, loss:0.25129374861717224
epoch: 180, loss:0.2505483329296112
epoch: 190, loss:0.23713809251785278
epoch: 200, loss:0.24216558039188385
epoch: 210, loss:0.22742848098278046
epoch: 220, loss:0.2271704226732254
epoch: 230, loss:0.2252519577741623
epoch: 240, loss:0.2193305790424347
epoch: 250, loss:0.21628950536251068
epoch: 260, loss:0.2177349030971527
epoch: 270, loss:0.2127998024225235
epoch: 280, loss:0.21056963503360748
epoch: 290, loss:0.2106262445449829
epoch: 300, loss:0.2119056135416031
epoch: 310, loss:0.21744157373905182
epoch: 320, loss:0.2134646326303482
epoch: 330, loss:0.20614351332187653
epoch: 340, loss:0.21047110855579376
epoch: 350, loss:0.20926110446453094
epoch: 360, loss:0.21647755801677704
epoch: 370, loss:0.21862877905368805
epoch: 380, loss:0.20591312646865845
epoch: 390, loss:0.2010645568370819
epoch: 400, loss:0.2050144225358963
epoch: 410, loss:0.1988362967967987
epoch: 420, loss:0.2004001885652542
epoch: 430, loss:0.1973736584186554
epoch: 440, loss:0.22496183216571808
epoch: 450, loss:0.19971159100532532
epoch: 460, loss:0.19692614674568176
epoch: 470, loss:0.19598537683486938
epoch: 480, loss:0.1948203295469284
epoch: 490, loss:0.19280420243740082
epoch: 500, loss:0.19459198415279388
epoch: 510, loss:0.19615624845027924
epoch: 520, loss:0.19460928440093994
epoch: 530, loss:0.19126252830028534
epoch: 540, loss:0.1938866376876831
epoch: 550, loss:0.19190427660942078
epoch: 560, loss:0.18747270107269287
epoch: 570, loss:0.18757325410842896
epoch: 580, loss:0.1988757699728012
epoch: 590, loss:0.19326147437095642
epoch: 600, loss:0.20158174633979797
epoch: 610, loss:0.18407107889652252
epoch: 620, loss:0.1833522766828537
epoch: 630, loss:0.18116755783557892
epoch: 640, loss:0.1835246980190277
epoch: 650, loss:0.18026278913021088
epoch: 660, loss:0.1771889179944992
epoch: 670, loss:0.1763589233160019
epoch: 680, loss:0.17852289974689484
epoch: 690, loss:0.17640434205532074
epoch: 700, loss:0.17527474462985992
epoch: 710, loss:0.18816079199314117
epoch: 720, loss:0.17597466707229614
epoch: 730, loss:0.17061764001846313
epoch: 740, loss:0.1716785877943039
epoch: 750, loss:0.16623708605766296
epoch: 760, loss:0.16821976006031036
epoch: 770, loss:0.163313090801239
epoch: 780, loss:0.1642996072769165
epoch: 790, loss:0.1681404858827591
epoch: 800, loss:0.16040602326393127
epoch: 810, loss:0.1622639298439026
epoch: 820, loss:0.16376478970050812
epoch: 830, loss:0.15491259098052979
epoch: 840, loss:0.15919749438762665
epoch: 850, loss:0.15590602159500122
epoch: 860, loss:0.15480470657348633
epoch: 870, loss:0.1528528481721878
epoch: 880, loss:0.1514061689376831
epoch: 890, loss:0.15681587159633636
epoch: 900, loss:0.1545121818780899
epoch: 910, loss:0.14895710349082947
epoch: 920, loss:0.14539872109889984
epoch: 930, loss:0.1600637137889862
epoch: 940, loss:0.1435800939798355
epoch: 950, loss:0.14283594489097595
epoch: 960, loss:0.14213287830352783
epoch: 970, loss:0.1424899399280548
epoch: 980, loss:0.14058111608028412
epoch: 990, loss:0.1395450234413147

训练模型在验证集上的损失
png