-
Notifications
You must be signed in to change notification settings - Fork 0
/
benchmark.py
176 lines (159 loc) · 5.51 KB
/
benchmark.py
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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import torch
import pandas as pd
import numpy as np
from tqdm import tqdm
def read_file(file_path):
df = pd.read_csv(file_path, nrows=5000)
df = df.dropna(axis = 'columns')
return df
class Preprocessor:
def __init__(self, cat_range=10):
self.cat_range = cat_range
self.properties = {}
def preprocess(self, df):
num_features = len(df.columns)-1
for col_name in df.columns:
col = df[col_name]
unique_val = np.unique(col).tolist()
col_range = len(unique_val)
if(type(col[0]) == str or col_range<self.cat_range):
self.properties[col_name]={'categorical':unique_val}
else:
stand_deviation = np.std(col)
mean = np.mean(col)
self.properties[col_name]={'numerical':{'mean':mean,'standard_deviation':stand_deviation}}
def transform(self,df):
data = {}
for col_name in self.properties.keys():
if(list(self.properties[col_name].keys())[0]=='categorical'):
vocab = self.properties[col_name]['categorical']
col_values = []
for row in df[col_name]:
col_values.append(vocab.index(row))
data[col_name]=col_values
else:
local_prop = self.properties[col_name]['numerical']
mean = local_prop['mean']
stand_deviation = local_prop['standard_deviation']
col_values = []
for row in df[col_name]:
col_values.append((row - mean)/stand_deviation)
data[col_name]=col_values
data = pd.DataFrame(data)
data.columns = self.properties.keys()
return data
class Model(torch.nn.Module):
def __init__(self, properties,target_var, sorted_columns, embedding_dim = 10, classification=True):
super(Model, self).__init__()
self.properties = properties
if(classification):
self.target_count = len(self.properties[target_var]['categorical'])
else:
self.target_count = 1
del self.properties[target_var]
self.count_num_features = 0
self.count_categorical_features = 0
self.categorical_features_embedding = []
self.sorted_columns = sorted_columns[:]
del self.sorted_columns[sorted_columns.index(target_var)]
self.target_var = target_var
self.embedding_dim = embedding_dim
self.preset()
self.output_linear = torch.nn.Linear(self.count_num_features+(self.count_categorical_features*self.embedding_dim), self.target_count)
def preset(self):
for col_name in self.sorted_columns:
if(list(self.properties[col_name].keys())[0]=='categorical'):
vocab = self.properties[col_name]['categorical']
self.count_categorical_features +=1
embedding = torch.nn.Embedding(num_embeddings = len(vocab), embedding_dim = self.embedding_dim)
self.categorical_features_embedding.append(embedding)
else:
self.count_num_features += 1
def forward(self, x_num, x_cat):
x = x_num
for index,col in enumerate(x_cat.transpose(0,1)):
embedding_layer = self.categorical_features_embedding[index]
embedding = embedding_layer(col)
x = torch.cat((x,embedding), 1)
x = self.output_linear(x)
return x
class TableDataset(torch.utils.data.Dataset):
def __init__(self, properties, data, target_var, classification = True):
self.properties = properties
self.columns = list(data.columns)
self.cat_columns = []
self.num_columns = []
self.data = data.values
self.target_var = target_var
self.classification=classification
self.preset()
def preset(self):
for col_name in self.columns:
if(list(self.properties[col_name].keys())[0]=='categorical'):
self.cat_columns.append(col_name)
else:
self.num_columns.append(col_name)
def __len__(self):
return self.data.shape[0]
def __getitem__(self, idx):
row = self.data[idx]
x_cat = []
x_num = []
if(self.classification):
y = int(row[self.columns.index(self.target_var)])
else:
y = row[self.columns.index(self.target_var)]
y = torch.Tensor([float(y)])
for v,col_name in zip(row,self.columns):
if(col_name==self.target_var):
continue
if(col_name in self.num_columns):
x_num.append(v)
else:
x_cat.append(v)
x_cat = torch.Tensor(x_cat)
x_cat = x_cat.to(torch.long)
x_num = torch.Tensor(x_num)
x_num = x_num.to(torch.float32)
return x_cat,x_num,y
def main(data, target_var, classification, lr):
preprocessor = Preprocessor()
preprocessor.preprocess(data)
data_transformed = preprocessor.transform(data)
dataset = TableDataset(preprocessor.properties, data_transformed, target_var, classification = classification)
dataloader = torch.utils.data.DataLoader(dataset, batch_size = 4, shuffle = True)
model = Model(preprocessor.properties, target_var, dataset.columns, classification = classification)
if(classification):
criterion = torch.nn.CrossEntropyLoss()
else:
criterion = torch.nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9)
for epoch in range(10):
running_loss =0.0
running_acc = 0.0
bar = tqdm(dataloader)
for i,(batch_x_cat, batch_x_num, batch_y) in enumerate(bar):
optimizer.zero_grad()
out = model(batch_x_num,batch_x_cat)
loss = criterion(out, batch_y)
loss.backward()
optimizer.step()
running_loss += loss.item()
if(classification):
pred = torch.argmax(out, dim = 1)
acc = pred == batch_y
acc = torch.mean(acc.float())
running_acc += acc.item()
bar.set_description(str({"epoch":epoch,"loss":round(running_loss/(i+1),3),"acc":round(running_acc/(i+1),3)}))
else:
bar.set_description(str({"epoch":epoch,"loss":round(running_loss/(i+1),3)}))
bar.close()
torch.save(model,"model.pt")
model = torch.load("model.pt")
print(model)
if __name__ == '__main__':
data = read_file("car.csv")
target_var = 'worth'
classification = True
lr = 0.001
main(data,target_var,classification,lr)