import numpy as np
import torch
import pandas as pd
from torch import nn
from torch.autograd import Variable
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import torch.nn.functional as F
from datetime import datetime
class Model(nn.Module):
def __init__(self, input_dim):
super(Model, self).__init__()
self.layer1 = nn.Linear(input_dim, 50)
self.layer2 = nn.Linear(50, 40)
self.layer3 = nn.Linear(40, 3)
def forward(self, x):
x = F.relu(self.layer1(x))
x = F.relu(self.layer2(x))
x = F.softmax(self.layer3(x)) # To check with the loss function
return x
# funkcja usuwająca wiersze zawierające platformę "Stadia"
def delete_stadia(games):
index_list = []
for i in range(0, len(games["platform"])):
try:
if games["platform"][i] == " Stadia":
index_list.append(i)
except:
continue
games.drop(index_list, inplace=True)
return games.reset_index()
# funkcja usuwająca wiersze zawierające "tbd" w kolumnie "user_review"
def delete_tbd(games):
index_list = []
for i in range(0, len(games["platform"])):
try:
if games["user_review"][i] == "tbd":
index_list.append(i)
except:
continue
games.drop(index_list, inplace=True)
return games.reset_index()
def delete_PC(games):
index_list = []
for i in range(0, len(games["platform"])):
try:
if games["platform"][i] == " PC":
index_list.append(i)
except:
continue
games.drop(index_list, inplace=True)
return games.reset_index()
# funkcja zmieniająca kolumnę "user_review" ze stringa na numeric
def user_review_to_numeric(games):
games["user_review"] = pd.to_numeric(games["user_review"])
return games
# funkcja normalizująca wartości w kolumnie "meta_score" i "user_review"
def normalization(games):
games['meta_score'] = games['meta_score'] / 100.0
games['user_review'] = games['user_review'] / 10.0
return games
# PlayStation - 0
# PlayStation 2 - 1
# PlayStation 3 - 2
# PlayStation 4 - 3
# PlayStation 5 - 4
# PlayStation Vita - 5
# Xbox - 6
# Xbox 360 - 7
# Xbox Series X - 8
# Nintendo 64 - 9
# GameCube - 10
# DS - 11
# 3DS - 12
# Wii - 13
# Wii U - 14
# Switch - 15
# PC - 16
# Dreamcast - 17
# Game Boy Advance - 18
# PSP - 19
# Xbox One - 20
# def platform_to_number(games):
# for i in range(0, len(games["platform"])):
#
# if games["platform"][i] == " PlayStation":
# games["platform"][i] = 0
# elif games["platform"][i] == " PlayStation 2":
# games["platform"][i] = 1
# elif games["platform"][i] == " PlayStation 3":
# games["platform"][i] = 2
# elif games["platform"][i] == " PlayStation 4":
# games["platform"][i] = 3
# elif games["platform"][i] == " PlayStation 5":
# games["platform"][i] = 4
# elif games["platform"][i] == " PlayStation Vita":
# games["platform"][i] = 5
# elif games["platform"][i] == " Xbox":
# games["platform"][i] = 6
# elif games["platform"][i] == " Xbox 360":
# games["platform"][i] = 7
# elif games["platform"][i] == " Xbox Series X":
# games["platform"][i] = 8
# elif games["platform"][i] == " Nintendo 64":
# games["platform"][i] = 9
# elif games["platform"][i] == " GameCube":
# games["platform"][i] = 10
# elif games["platform"][i] == " DS":
# games["platform"][i] = 11
# elif games["platform"][i] == " 3DS":
# games["platform"][i] = 12
# elif games["platform"][i] == " Wii":
# games["platform"][i] = 13
# elif games["platform"][i] == " Wii U":
# games["platform"][i] = 14
# elif games["platform"][i] == " Switch":
# games["platform"][i] = 15
# elif games["platform"][i] == " PC":
# games["platform"][i] = 16
# elif games["platform"][i] == " Dreamcast":
# games["platform"][i] = 17
# elif games["platform"][i] == " Game Boy Advance":
# games["platform"][i] = 18
# elif games["platform"][i] == " PSP":
# games["platform"][i] = 19
# elif games["platform"][i] == " Xbox One":
# games["platform"][i] = 20
#
# return games
# old - 0
# mid - 1
# new - 2
def platform_to_number(games):
for i in range(0, len(games["platform"])):
if games["platform"][i] == " PlayStation":
games["platform"][i] = 0
elif games["platform"][i] == " PlayStation 2":
games["platform"][i] = 0
elif games["platform"][i] == " PlayStation 3":
games["platform"][i] = 1
elif games["platform"][i] == " PlayStation 4":
games["platform"][i] = 2
elif games["platform"][i] == " PlayStation 5":
games["platform"][i] = 2
elif games["platform"][i] == " PlayStation Vita":
games["platform"][i] = 1
elif games["platform"][i] == " Xbox":
games["platform"][i] = 0
elif games["platform"][i] == " Xbox 360":
games["platform"][i] = 1
elif games["platform"][i] == " Xbox Series X":
games["platform"][i] = 2
elif games["platform"][i] == " Nintendo 64":
games["platform"][i] = 0
elif games["platform"][i] == " GameCube":
games["platform"][i] = 0
elif games["platform"][i] == " DS":
games["platform"][i] = 0
elif games["platform"][i] == " 3DS":
games["platform"][i] = 1
elif games["platform"][i] == " Wii":
games["platform"][i] = 0
elif games["platform"][i] == " Wii U":
games["platform"][i] = 1
elif games["platform"][i] == " Switch":
games["platform"][i] = 2
elif games["platform"][i] == " PC":
dt = datetime.strptime(games["release_date"][i], '%B %d, %Y')
if (dt.year == 1995 or dt.year == 1996 or dt.year == 1997 or dt.year == 1998
or dt.year == 1999 or dt.year == 2000 or dt.year == 2001 or dt.year == 2002
or dt.year == 2003 or dt.year == 2004 or dt.year == 2005):
games["platform"][i] = 0
if (dt.year == 2006 or dt.year == 2007 or dt.year == 2008 or dt.year == 2009
or dt.year == 2010 or dt.year == 2011 or dt.year == 2012 or dt.year == 2013
or dt.year == 2014 or dt.year == 2015 or dt.year == 2016):
games["platform"][i] = 1
if (dt.year == 2017 or dt.year == 2018 or dt.year == 2019
or dt.year == 2020 or dt.year == 2021):
games["platform"][i] = 2
# games["platform"][i] = 0
elif games["platform"][i] == " Dreamcast":
games["platform"][i] = 0
elif games["platform"][i] == " Game Boy Advance":
games["platform"][i] = 0
elif games["platform"][i] == " PSP":
games["platform"][i] = 1
elif games["platform"][i] == " Xbox One":
games["platform"][i] = 2
return games
games = pd.read_csv('/dane/all_games.csv', sep=',')
games = platform_to_number(games)
games = delete_stadia(games)
games = delete_tbd(games)
games = user_review_to_numeric(games)
games = normalization(games)
games.drop(['level_0', 'index'], axis='columns', inplace=True)
labels_g = pd.DataFrame(games["platform"], dtype=np.int64)
labels_g = labels_g.to_numpy()
features_g = {'meta_score': games['meta_score'],
'user_review': games['user_review']}
features_g = pd.DataFrame(features_g, dtype=np.float64)
features_g = features_g.to_numpy()
features_train_g, features_test_g, labels_train_g, labels_test_g = train_test_split(features_g,
labels_g,
random_state=1,
shuffle=True)
# Training
model = Model(features_train_g.shape[1])
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
loss_fn = nn.CrossEntropyLoss()
epochs = 1000
def print_(loss):
print ("The loss calculated: ", loss)
# Not using dataloader
x_train, y_train = Variable(torch.from_numpy(features_train_g)).float(), Variable(torch.from_numpy(labels_train_g)).long()
for epoch in range(1, epochs + 1):
print("Epoch #", epoch)
y_pred = model(x_train)
loss = loss_fn(y_pred, y_train.squeeze(-1))
print_(loss.item())
# Zero gradients
optimizer.zero_grad()
loss.backward() # Gradients
optimizer.step() # Update
# Prediction
x_test = Variable(torch.from_numpy(features_test_g)).float()
pred = model(x_test)
pred = pred.detach().numpy()
print("The accuracy is", accuracy_score(labels_test_g, np.argmax(pred, axis=1)))
pred = pd.DataFrame(pred)
pred.to_csv('result.csv')