Bitcoin hash-rateは二次攻撃を予測する

# This Python 3 environment comes with many helpful analytics libraries installed
# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python
# For example, here's several helpful packages to load

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)

# Input data files are available in the read-only "../input/" directory
# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory

import os
for dirname, _, filenames in os.walk('/kaggle/input'):
    for filename in filenames:
        print(os.path.join(dirname, filename))

# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using "Save & Run All" 
# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session

データのロード

train = pd.read_csv('/kaggle/input/sejongai-hashrate/train.csv')
test = pd.read_csv('/kaggle/input/sejongai-hashrate/test.csv')
sample = pd.read_csv('/kaggle/input/sejongai-hashrate/submit_sample.csv')

モジュールインポートとGPUの使用

import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.optim as optim
import random
from sklearn.preprocessing import StandardScaler

device ='cuda' if torch.cuda.is_available() else 'cpu'

random.seed(777)
torch.manual_seed(777)
if device == 'cuda' :
    torch.cuda.manual_seed_all(777)

device

データグループ

x_train = train.iloc[:,1:-1]
# year
x_train['Timestamp'].str.split(' ').str[0].str.split('/').str[2].astype(int)

# month
x_train['Timestamp'].str.split('/').str[0].astype(int)

# date
x_train['Timestamp'].str.split(' ').str[0].str.split('/').str[1].astype(int)

x_train['month'] = x_train['Timestamp'].str.split('/').str[0].astype(int)
x_test = test.iloc[:,1:]
x_test['month'] = x_test['Timestamp'].str.split('/').str[0].astype(int)
y_train = train.iloc[:,-1]

x_train['year'] = x_train['Timestamp'].str.split(' ').str[0].str.split('/').str[2].astype(int)
x_test['year'] = x_test['Timestamp'].str.split(' ').str[0].str.split('/').str[2].astype(int)

x_train['date'] = x_train['Timestamp'].str.split(' ').str[0].str.split('/').str[1].astype(int)
x_test['date'] = x_test['Timestamp'].str.split(' ').str[0].str.split('/').str[1].astype(int)

x_train = x_train.drop(['Timestamp'],axis=1)
x_test = x_test.drop(['Timestamp'],axis=1)
# x_train = x_train.drop(['n-unique-addresses'], axis=1)
# x_test = x_test.drop(['n-unique-addresses'], axis=1)

sc = StandardScaler()
x_train = sc.fit_transform(x_train)
x_test = sc.transform(x_test)

x_train = torch.FloatTensor(x_train).to(device)
x_test = torch.FloatTensor(x_test).to(device)
y_train = torch.FloatTensor(y_train).reshape(-1,1).to(device)

データ検証

# train data
print(x_train[:5])
print(x_train.shape)
print(y_train[:5])
print(y_train.shape)

# test data
print(x_test[:3])

モデル定義

class NN(torch.nn.Module):
    def __init__(self):
        super(NN,self).__init__()
        
        self.linear1 = nn.Linear(9,512,bias=True)
        self.linear2 = nn.Linear(512,256,bias=True)
        self.linear3 = nn.Linear(256,512,bias=True)
        self.linear4 = nn.Linear(512,64,bias=True)
        self.linear5 = nn.Linear(64,32,bias=True)
        self.linear6 = nn.Linear(32,1,bias=True)
        self.relu = nn.ReLU()
        
        torch.nn.init.orthogonal_(self.linear1.weight)
        torch.nn.init.orthogonal_(self.linear2.weight)
        torch.nn.init.orthogonal_(self.linear3.weight)
        torch.nn.init.orthogonal_(self.linear4.weight)
        torch.nn.init.orthogonal_(self.linear5.weight)
        torch.nn.init.orthogonal_(self.linear6.weight)
        
    def forward(self,x):
        out = self.linear1(x)
        out = self.relu(out)
        out = self.linear2(out)
        out = self.relu(out)
        out = self.linear3(out)
        out = self.relu(out)
        out = self.linear4(out)
        out = self.relu(out)
        out = self.linear5(out)
        out = self.relu(out)
        out = self.linear6(out)
        return out

学習パラメータの設定

model = NN().to(device)
optimizer = optim.Adam(model.parameters(), lr=15e-5)
loss = nn.MSELoss().to(device)
epochs = 1700
model

学習モデル

model.train()
plt_los = []
train_total_batch = len(x_train)
for epoch in range(epochs+1) : 
    avg_cost = 0
    model.train()

    hypothesis = model(x_train) 
    cost = loss(hypothesis,y_train) 
    
    optimizer.zero_grad() 
    cost.backward()
    optimizer.step() 
    avg_cost += cost / train_total_batch
    plt_los.append([avg_cost.item()])
    if epoch%100==0:
        print('Epoch : {}, Cost : {}'.format(epoch, avg_cost.item()))

Plot

import matplotlib.pyplot as plt

def plot(loss_list: list, ylim=None, title=None) -> None:
    bn = [i[0] for i in loss_list]

    plt.figure(figsize=(10, 10))
    plt.plot(bn, label='avg_cost')
    if ylim:
        plt.ylim(ylim)

    if title:
        plt.title(title)
    plt.legend()
    plt.grid('on')
    plt.show()
    
plot(plt_los , [0.0, 1.0], title='Loss at Epoch')
print(avg_cost.item())

予測のエクスポートとコミット

model.eval()
with torch.no_grad():
    y_pred = model(x_test)
sample['hash-rate'] = y_pred.cpu().numpy()
sample.to_csv('submit_sample.csv',index=False)
sample

マザーボードの結果