In [1]:
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
#data preparation
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.impute import SimpleImputer
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.model_selection import train_test_split
#regression models
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.svm import SVR
#evaluation
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import r2_score
In [2]:
df = pd.read_csv(
"Alien Data Set.csv", )
print(df.info())
df = df[df["Diameter"]<100]
# print(df.head())
# print(df["Age"].nunique())
fig = plt.figure(figsize=(20,20))
for col in range(len(df.columns)):
fig.add_subplot(6,3,col+1)
sns.boxplot(x=df.iloc[:,col])
plt.show()
df.head()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 3893 entries, 0 to 3892 Data columns (total 8 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 Sex 3893 non-null object 1 Length 3893 non-null float64 2 Diameter 3893 non-null float64 3 Height 3893 non-null float64 4 Shell Mass 3893 non-null float64 5 Visceral Fat Mass 3893 non-null float64 6 Body Fat Mass 3893 non-null float64 7 Age 3893 non-null int64 dtypes: float64(6), int64(1), object(1) memory usage: 243.4+ KB None
Out[2]:
| Sex | Length | Diameter | Height | Shell Mass | Visceral Fat Mass | Body Fat Mass | Age | |
|---|---|---|---|---|---|---|---|---|
| 0 | F | 1.4375 | 1.1750 | 0.4125 | 6.747181 | 5.584852 | 12.332033 | 9 |
| 1 | M | 0.8875 | 0.6500 | 0.2125 | 1.559222 | 1.374951 | 2.296310 | 6 |
| 2 | I | 1.0375 | 0.7750 | 0.2500 | 2.764076 | 1.601747 | 3.231843 | 6 |
| 3 | F | 1.1750 | 0.8875 | 0.2500 | 5.244657 | 2.282135 | 4.748541 | 10 |
| 4 | I | 0.8875 | 0.6625 | 0.2125 | 1.700970 | 1.488349 | 3.458639 | 6 |
In [3]:
# fig = plt.figure(figsize=(20,20))
# for col in range(2):
# fig.add_subplot(6,3,col+1)
# sns.displot(df.iloc[:,col])
# plt.show()
sns.heatmap(df.iloc[:,1:-1].corr(),annot=True, cmap='coolwarm')
Out[3]:
<Axes: >
Check why there are values of age less than 0. It seems not a data mistake, as the other data entries look normal. Maybe alien age starts from -1.
In [15]:
df['Age'].value_counts(normalize=True)
df_negative_age = df[df['Age']<0]
df_negative_age
# trying to add another total mass column
# df["total mass"]= df["Shell Mass"]+df["Visceral Fat Mass"]+df["Body Fat Mass"]
# see how it affects the performance by dropping some columns
# df = df.drop(columns=["Body Fat Mass"])
# plt.hist(df['Diameter'], bins=30, color='blue', edgecolor='black',density=True)
sns.histplot(df['Age'], kde=True)
plt.xlabel('age')
plt.ylabel('probability')
plt.title('alen age distribution')
Out[15]:
Text(0.5, 1.0, 'alen age distribution')
data transformation¶
In [5]:
X=df.drop(columns=['Age'],axis=1)
y = df['Age'].astype(float)
num_features = X.select_dtypes(exclude="object").columns
cat_features = X.select_dtypes(include="object").columns
num_pipeline= Pipeline(
steps=[
("imputer",SimpleImputer(strategy="median")),
("scaler",StandardScaler())
]
)
cat_pipeline=Pipeline(
steps=[
("imputer",SimpleImputer(strategy="most_frequent")),
("one_hot_encoder",OneHotEncoder()),
("scaler",StandardScaler(with_mean=False))
]
)
preprocessor=ColumnTransformer(
[
("num_pipeline",num_pipeline,num_features),
("cat_pipelines",cat_pipeline,cat_features)
]
)
X = preprocessor.fit_transform(X)
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2,random_state=42)
select 5 regression models and compare their performances, only the best will be used¶
In [6]:
models = {
"Linear Regression": LinearRegression(),
"Decision Tree": DecisionTreeRegressor(),
"Random Forest": RandomForestRegressor(),
"Gradient Boosting": GradientBoostingRegressor(),
"Support Vector": SVR(),
}
def evaluate_model(true, predicted):
mae = mean_absolute_error(true, predicted)
score = r2_score(true, predicted)
return (mae, score)
model_list = []
mae_list =[]
for i in range(len(list(models))):
model = list(models.values())[i]
model.fit(X_train, y_train) # Train model
# Make predictions
y_train_pred = model.predict(X_train)
y_test_pred = model.predict(X_test)
model_train_accuracy, r2_train = evaluate_model(y_train, y_train_pred)
model_test_accuracy, r2_test = evaluate_model(y_test, y_test_pred)
print(list(models.keys())[i])
print('Model performance for Training set')
# print("- accuracy: {:.4f}".format(model_train_accuracy))
# print("- r2: {:.4f}".format(r2_train))
# print('----------------------------------')
print('Model performance for Testing set')
print("- accuracy: {:.4f}".format(model_test_accuracy))
print("- r2: {:.4f}".format(r2_test))
model_list.append(model)
mae_list.append(model_test_accuracy)
print('='*35)
Linear Regression Model performance for Training set Model performance for Testing set - accuracy: 1.6480 - r2: 0.4604 =================================== Decision Tree Model performance for Training set Model performance for Testing set - accuracy: 2.1964 - r2: 0.0201 =================================== Random Forest Model performance for Training set Model performance for Testing set - accuracy: 1.6459 - r2: 0.4710 =================================== Gradient Boosting Model performance for Training set Model performance for Testing set - accuracy: 1.5802 - r2: 0.5017 =================================== Support Vector Model performance for Training set Model performance for Testing set - accuracy: 1.5495 - r2: 0.4833 ===================================
best model is linear regression¶
In [12]:
best_model= model_list[-2]
error = abs(best_model.predict(X_test)-y_test)
# plt.hist(error, bins=30, color='blue', edgecolor='black',density=True)
sns.histplot(error, kde=True)
plt.xlabel('mean absolute error')
plt.ylabel('density')
plt.title('Model prediction performance')
n=len(error)
print(len(error[error<0.5])/n)
print(len(error[error<1])/n)
print(len(error[error<2])/n)
print(len(error[error<3])/n)
0.23491655969191272 0.47753530166880614 0.7355584082156611 0.8703465982028241
In [14]:
fig, ax1 = plt.subplots()
# Plot the histogram on the primary axis
ax1.hist(df['Age'], bins=30, color='blue', edgecolor='black', density=True)
ax1.set_xlabel('Age')
ax1.set_ylabel('Age distribution density', color='blue')
ax1.set_title('Alien Age Distribution vs. model performance')
# Create secondary axis
ax2 = ax1.twinx()
# Plot the scatter plot on the secondary axis
ax2.scatter(y_test, error, color='red', label='Accuracy',s=4)
ax2.set_ylabel('mean absolute error', color='red')
# Show the plot
plt.show()
