In this part we will assess ML model
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
sns.set_theme(style="whitegrid")
data = pd.read_csv('../dataset/extracted/feature_extracted.csv')
# data=pd.read_csv('../dataset/processed/cleanedData.csv')
data.head(2)
| Movie_Title | Movie_ID | Budget | Cast | Crew | Studios | Genre | Keywords | Languages | Countries | ... | HasTopCrew | NumStudios | StudioRank | NumTopStudios | HasTopStudio | isUnitedStates | CountryRank | IsEnglish | keywordRank | cerRank | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | Star Wars: Episode VII - The Force Awakens | 2488496 | 245000000 | ['Daisy Ridley', 'John Boyega', 'Oscar Isaac',... | ['Lawrence Kasdan', 'Michael Arndt', 'J.J. Abr... | ['Lucasfilm', 'Bad Robot', 'Truenorth Producti... | ['Action', 'Adventure', 'Sci-Fi'] | ['reboot', 'sanitation employee', 'remake', 'c... | ['English'] | ['United States'] | ... | 1 | 3 | 1324 | 2 | 1 | 1 | 6 | 1 | 82.6 | 5 |
| 1 | Frozen II | 4520988 | 150000000 | ['Kristen Bell', 'Idina Menzel', 'Josh Gad', '... | ['Jennifer Lee', 'Hans Christian Andersen', 'C... | ['Walt Disney Animation Studios', 'Walt Disney... | ['Animation', 'Adventure', 'Comedy', 'Family',... | ['autumn', 'anthropomorphic snowman', 'princes... | ['English'] | ['United States'] | ... | 1 | 2 | 1126 | 2 | 1 | 1 | 6 | 1 | 142.0 | 4 |
2 rows × 38 columns
Drop str column
for index in data.columns:
if type(data[index][0]) == str:
data = data.drop(index, axis=1)
# Drop Movie_ID, Rating, Rating_Count
data.drop(['Movie_ID', 'Rating', 'Rating_Count', 'Release_Day', 'Release_Month'], axis=1, inplace=True)
data.head(2)
| Budget | Runtime | Gross_worldwide | Release_Year | GenreRank | IsAdventure | SpecialMonth | CastsRank | NumLeadActors | HasTop50Actors | ... | HasTopCrew | NumStudios | StudioRank | NumTopStudios | HasTopStudio | isUnitedStates | CountryRank | IsEnglish | keywordRank | cerRank | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 245000000 | 138.0 | 2069521700 | 2015 | 19.666667 | 1 | 1 | 58314 | 8 | 1 | ... | 1 | 3 | 1324 | 2 | 1 | 1 | 6 | 1 | 82.6 | 5 |
| 1 | 150000000 | 103.0 | 1450026933 | 2019 | 17.333333 | 1 | 1 | 41136 | 2 | 0 | ... | 1 | 2 | 1126 | 2 | 1 | 1 | 6 | 1 | 142.0 | 4 |
2 rows × 23 columns
data.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 8752 entries, 0 to 8751 Data columns (total 23 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 Budget 8752 non-null int64 1 Runtime 8752 non-null float64 2 Gross_worldwide 8752 non-null int64 3 Release_Year 8752 non-null int64 4 GenreRank 8752 non-null float64 5 IsAdventure 8752 non-null int64 6 SpecialMonth 8752 non-null int64 7 CastsRank 8752 non-null int64 8 NumLeadActors 8752 non-null int64 9 HasTop50Actors 8752 non-null int64 10 NumCrews 8752 non-null int64 11 crewsTeamRank 8752 non-null int64 12 NumTopCrew 8752 non-null int64 13 HasTopCrew 8752 non-null int64 14 NumStudios 8752 non-null int64 15 StudioRank 8752 non-null int64 16 NumTopStudios 8752 non-null int64 17 HasTopStudio 8752 non-null int64 18 isUnitedStates 8752 non-null int64 19 CountryRank 8752 non-null int64 20 IsEnglish 8752 non-null int64 21 keywordRank 8752 non-null float64 22 cerRank 8752 non-null int64 dtypes: float64(3), int64(20) memory usage: 1.5 MB
Correlation Plot
plt.figure(figsize=(30, 30))
sns.heatmap(data.corr(), annot=True)
plt.show()
Top correlation with Gross_worldwide
data.corr()['Gross_worldwide'].sort_values(ascending=False)
Gross_worldwide 1.000000 Budget 0.741255 NumTopCrew 0.620654 NumLeadActors 0.593344 crewsTeamRank 0.550060 HasTopCrew 0.543905 NumTopStudios 0.529891 CastsRank 0.526255 HasTopStudio 0.503802 StudioRank 0.473315 IsAdventure 0.366146 HasTop50Actors 0.363827 GenreRank 0.325225 Runtime 0.210073 Release_Year 0.209071 cerRank 0.199208 NumCrews 0.190015 keywordRank 0.162205 CountryRank 0.153409 SpecialMonth 0.141635 NumStudios 0.133288 isUnitedStates 0.128513 IsEnglish 0.075937 Name: Gross_worldwide, dtype: float64
import numpy as np
from sklearn import linear_model
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
data.columns
Index(['Budget', 'Runtime', 'Gross_worldwide', 'Release_Year', 'GenreRank',
'IsAdventure', 'SpecialMonth', 'CastsRank', 'NumLeadActors',
'HasTop50Actors', 'NumCrews', 'crewsTeamRank', 'NumTopCrew',
'HasTopCrew', 'NumStudios', 'StudioRank', 'NumTopStudios',
'HasTopStudio', 'isUnitedStates', 'CountryRank', 'IsEnglish',
'keywordRank', 'cerRank'],
dtype='object')
testScore = 0
trainScore = 0
mae_val = 0
mae_train = 0
loop = 1000
for i in range(loop):
lin = linear_model.LinearRegression()
train, test = train_test_split(data, test_size=0.3)
target = ['Gross_worldwide']
feature = ['Budget', 'Runtime', 'Release_Year']
x_train = train[feature]
y_train = train[target]
x_test = test[feature]
y_test = test[target]
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore += lin.score(x_test, y_test)
trainScore += lin.score(x_train, y_train)
mae_train += mean_absolute_error(y_train, lin.predict(x_train))
mae_val += mean_absolute_error(y_test, y_predicted)
else:
testScore /= loop
trainScore /= loop
mae_val /= loop
mae_train /= loop
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE Train: ", mae_train)
print("MAE Test: ", mae_val)
Average train score: 0.55 Average test score: 0.55 MAE Train: 51511296.55432569 MAE Test: 51453703.08499506
Mean MAE of test set with 1000 loops is 52M$ ~ 1200 Tỷ VND
testScore = 0
trainScore = 0
mae_val = 0
mae_train = 0
loop = 100
for i in range(loop):
lin = linear_model.LinearRegression()
train, test = train_test_split(data, test_size=0.3)
target = ['Gross_worldwide']
x_train = train.drop('Gross_worldwide', axis=1)
y_train = train[target]
x_test = test.drop('Gross_worldwide', axis=1)
y_test = test[target]
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore += lin.score(x_test, y_test)
trainScore += lin.score(x_train, y_train)
mae_train += mean_absolute_error(y_train, lin.predict(x_train))
mae_val += mean_absolute_error(y_test, y_predicted)
else:
testScore /= loop
trainScore /= loop
mae_val /= loop
mae_train /= loop
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE Train: ", mae_train)
print("MAE Test: ", mae_val)
Average train score: 0.69 Average test score: 0.68 MAE Train: 45876763.969360344 MAE Test: 46112852.09602557
Mean MAE of test set with 1000 loops is 46M$ ~ 1050 Tỷ VND
from sklearn.ensemble import RandomForestRegressor
from sklearn.ensemble import RandomForestRegressor
lin = RandomForestRegressor(n_estimators=1000)
train, test = train_test_split(data, test_size=0.3)
target = ['Gross_worldwide']
x_train = train.drop('Gross_worldwide', axis=1)
y_train = train[target]
x_test = test.drop('Gross_worldwide', axis=1)
y_test = test[target]
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore = lin.score(x_test, y_test)
trainScore = lin.score(x_train, y_train)
mae_val = mean_absolute_error(y_test, y_predicted)
mae_val_train = mean_absolute_error(y_train, lin.predict(x_train))
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE train: ", mae_val_train)
print("MAE test: ", mae_val)
Mean MAE of test 38M$ ~ 870 Tỷ Vietnam Dong
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
sns.set_theme(style="whitegrid")
data = pd.read_csv('../dataset/processed/cleaned_data.csv')
data.head(2)
from sklearn import linear_model
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error
from FeatureExtractor import BORFeatureExtractor
from sklearn.preprocessing import MinMaxScaler
import ast
cols = ['Cast', 'Genre', 'Studios', 'ListOfCertificate', 'Keywords', 'Languages', 'Countries', 'Crew']
for col in cols:
data[col] = data[col].apply(ast.literal_eval)
data.columns
testScore = 0
trainScore = 0
mae_val = 0
mae_train = 0
loop = 10
for i in range(loop):
lin = linear_model.LinearRegression()
train, test = train_test_split(data, test_size=0.2)
train.reset_index(drop=True, inplace=True)
test.reset_index(drop=True, inplace=True)
fe = BORFeatureExtractor()
fe.fit(train)
fe.extract(train)
fe.extract(test)
target = ['Gross_worldwide']
x_train = train.drop('Gross_worldwide', axis=1)
y_train = train[target]
x_test = test.drop('Gross_worldwide', axis=1)
y_test = test[target]
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore += lin.score(x_test, y_test)
trainScore += lin.score(x_train, y_train)
mae_train += mean_absolute_error(y_train, lin.predict(x_train))
mae_val += mean_absolute_error(y_test, y_predicted)
else:
testScore /= loop
trainScore /= loop
mae_val /= loop
mae_train /= loop
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE Train: ", mae_train)
print("MAE Test: ", mae_val)
Since it takes time to extract feature while running so we run only 10 time.
The result is MAE on test is 48M $
Now we will see Linear Regression give how much coeficient on Data
testScore = 0
trainScore = 0
mae_val = 0
mae_train = 0
loop = 1
for i in range(loop):
lin = linear_model.LinearRegression()
train, test = train_test_split(data, test_size=0.2)
train.reset_index(drop=True, inplace=True)
test.reset_index(drop=True, inplace=True)
fe = BORFeatureExtractor()
scale = MinMaxScaler()
fe.fit(train)
fe.extract(train)
fe.extract(test)
target = ['Gross_worldwide']
x_train = train.drop('Gross_worldwide', axis=1)
y_train = train[target]
x_test = test.drop('Gross_worldwide', axis=1)
y_test = test[target]
columns = list(x_train.columns)
scale = scale.fit(x_train)
x_train = scale.transform(x_train)
x_test = scale.transform(x_test)
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore += lin.score(x_test, y_test)
trainScore += lin.score(x_train, y_train)
mae_train += mean_absolute_error(y_train, lin.predict(x_train))
mae_val += mean_absolute_error(y_test, y_predicted)
else:
testScore /= loop
trainScore /= loop
mae_val /= loop
mae_train /= loop
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE Train: ", mae_train)
print("MAE Test: ", mae_val)
Now conduct a coefficient table for each attributes to see what are the best predictors
coef_df = pd.DataFrame(np.abs(lin.coef_).T, columns=['Coef'], index=columns)
coef_df.sort_values(by='Coef', ascending=False)
from sklearn.ensemble import RandomForestRegressor
lin = RandomForestRegressor(n_estimators=100)
train, test = train_test_split(data, test_size=0.3)
## Extract feature
train.reset_index(drop=True, inplace=True)
test.reset_index(drop=True, inplace=True)
fe = BORFeatureExtractor()
fe.fit(train)
fe.extract(train)
fe.extract(test)
##
target = ['Gross_worldwide']
x_train = train.drop('Gross_worldwide', axis=1)
y_train = train[target]
x_test = test.drop('Gross_worldwide', axis=1)
y_test = test[target]
lin.fit(x_train, y_train)
y_predicted = lin.predict(x_test)
testScore = lin.score(x_test, y_test)
trainScore = lin.score(x_train, y_train)
mae_val = mean_absolute_error(y_test, y_predicted)
mae_val_train = mean_absolute_error(y_train, lin.predict(x_train))
print("Average train score: %.2f" % trainScore)
print("Average test score: %.2f" % testScore)
print("MAE train: ", mae_val_train)
print("MAE test: ", mae_val)
We run only 1 time with 1000 estimators.
The result is MAE on test is 43M $