app.py

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
import mlflow
import mlflow.sklearn
from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score


from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.model_selection import GridSearchCV
import mlflow
import mlflow.sklearn
import joblib

# Load dataset
df = pd.read_csv('ds_salaries.csv')

# EDA
print(df.head())
print(df.info())
print(df.describe())

# Visualizations
sns.pairplot(df)
plt.show()

# Handle missing values
# Fill numeric columns with mean and categorical columns with mode
numeric_cols = df.select_dtypes(include=['float64', 'int64']).columns
categorical_cols = df.select_dtypes(include=['object']).columns

df[numeric_cols] = df[numeric_cols].apply(lambda x: x.fillna(x.mean()))
df[categorical_cols] = df[categorical_cols].apply(lambda x: x.fillna(x.mode()[0]))

# Drop the salary_currency column as it's not needed for prediction
df = df.drop(columns=['salary_currency'])

# Encode categorical variables
categorical_columns = ['experience_level', 'employment_type', 'job_title', 'employee_residence', 'company_location', 'company_size']
df = pd.get_dummies(df, columns=categorical_columns, drop_first=True)

# Define features and target variable
X = df.drop(['salary', 'salary_in_usd'], axis=1)
y = df['salary_in_usd']

# Split the data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Model training and experiment tracking with MLflow
models = {
    'Linear Regression': LinearRegression(),
    'Decision Tree': DecisionTreeRegressor(),
    'Random Forest': RandomForestRegressor(),
    'Gradient Boosting': GradientBoostingRegressor()
}

mlflow.set_experiment('Data Science Salaries Prediction')

for model_name, model in models.items():
    with mlflow.start_run(run_name=model_name):
        model.fit(X_train, y_train)
        y_pred = model.predict(X_test)
        
        # Log model
        mlflow.sklearn.log_model(model, model_name)
        
        # Log metrics
        mlflow.log_metric('RMSE', mean_squared_error(y_test, y_pred, squared=False))
        mlflow.log_metric('MAE', mean_absolute_error(y_test, y_pred))
        mlflow.log_metric('R2', r2_score(y_test, y_pred))

# Hyperparameter tuning for the best model (e.g., Random Forest)
from sklearn.model_selection import GridSearchCV

param_grid = {
    'n_estimators': [100, 200, 300],
    'max_depth': [None, 10, 20, 30]
}

grid_search = GridSearchCV(RandomForestRegressor(), param_grid, cv=3, scoring='r2')
grid_search.fit(X_train, y_train)

# Log the best model and parameters
with mlflow.start_run(run_name='Optimized Random Forest'):
    mlflow.sklearn.log_model(grid_search.best_estimator_, 'Random Forest')
    mlflow.log_params(grid_search.best_params_)
    mlflow.log_metric('Best R2', grid_search.best_score_)














# Model training and experiment tracking with MLflow
models = {
    'Linear Regression': LinearRegression(),
    'Decision Tree': DecisionTreeRegressor(),
    'Random Forest': RandomForestRegressor(),
    'Gradient Boosting': GradientBoostingRegressor()
}

mlflow.set_experiment('Data Science Salaries Prediction')

for model_name, model in models.items():
    with mlflow.start_run(run_name=model_name):
        model.fit(X_train, y_train)
        y_pred = model.predict(X_test)
        
        # Log model
        mlflow.sklearn.log_model(model, model_name)
        
        # Log metrics
        mlflow.log_metric('RMSE', mean_squared_error(y_test, y_pred, squared=False))
        mlflow.log_metric('MAE', mean_absolute_error(y_test, y_pred))
        mlflow.log_metric('R2', r2_score(y_test, y_pred))

# Hyperparameter tuning for the best model (e.g., Random Forest)
param_grid = {
    'n_estimators': [100, 200, 300],
    'max_depth': [None, 10, 20, 30]
}

grid_search = GridSearchCV(RandomForestRegressor(), param_grid, cv=3, scoring='r2')
grid_search.fit(X_train, y_train)

# Log the best model and parameters
with mlflow.start_run(run_name='Optimized Random Forest'):
    mlflow.sklearn.log_model(grid_search.best_estimator_, 'Random Forest')
    mlflow.log_params(grid_search.best_params_)
    mlflow.log_metric('Best R2', grid_search.best_score_)

# Save the best model
joblib.dump(grid_search.best_estimator_, 'best_model.pkl')


















import streamlit as st
import pandas as pd
import joblib

# Load the original dataset to get unique values for dropdowns
df_original = pd.read_csv('ds_salaries.csv')

# Load the best model
best_model = joblib.load('best_model.pkl')

# Streamlit app
st.title('Data Science Salary Predictor')

# Input features
experience_level = st.selectbox('Experience Level', df_original['experience_level'].unique())
employment_type = st.selectbox('Employment Type', df_original['employment_type'].unique())
job_title = st.selectbox('Job Title', df_original['job_title'].unique())
employee_residence = st.selectbox('Employee Residence', df_original['employee_residence'].unique())
remote_ratio = st.selectbox('Remote Ratio', df_original['remote_ratio'].unique())
company_location = st.selectbox('Company Location', df_original['company_location'].unique())
company_size = st.selectbox('Company Size', df_original['company_size'].unique())

# Predict salary
input_data = pd.DataFrame({
    'work_year': [2023],
    'experience_level': [experience_level],
    'employment_type': [employment_type],
    'job_title': [job_title],
    'employee_residence': [employee_residence],
    'remote_ratio': [remote_ratio],
    'company_location': [company_location],
    'company_size': [company_size]
})

# Encode categorical variables
categorical_columns = ['experience_level', 'employment_type', 'job_title', 'employee_residence', 'company_location', 'company_size']
input_data = pd.get_dummies(input_data, columns=categorical_columns, drop_first=True)

# Align input data with training data columns
input_data = input_data.reindex(columns=X_train.columns, fill_value=0)

# Predict the salary
salary_prediction = best_model.predict(input_data)[0]
st.write(f'Predicted Salary: ${salary_prediction:.2f}')