Speed compare between CPU and GPU xgboost

nrow-time-xgboost.py

# Defining the dictionaries
cpu_dict = {
    'objective': 'reg:squarederror'
}

gpu_dict = {
    'objective': 'reg:squarederror',
    'tree_method': 'gpu_hist'
}

# Number of rows to train on 
nrows = [
    10, 
    100, 
    200, 
    3000, 
    5000, 
    10000, 
    20000, 
    50000, 
    100000, 
    300000
]
cpu_speeds = []
gpu_speeds = []

for nrow in nrows: 
    # Sampling random rows 
    sample = d.sample(nrow)
    
    # Creating X and Y 
    X = pd.get_dummies(sample[features])
    Y = sample['Sales']
    
    # Initiating the model objects
    cpu = xgb.XGBRegressor(**cpu_dict)
    gpu = xgb.XGBRegressor(**gpu_dict)
    
    # Training on cpu 
    start = time.time()
    cpu.fit(X, Y)
    cpu_speed = time.time() - start
    
    # Training on gpu
    start = time.time()
    gpu.fit(X, Y)
    gpu_speed = time.time() - start
    
    # Appending the speed results 
    cpu_speeds.append(cpu_speed)
    gpu_speeds.append(gpu_speed)

# Creating a dataframe 
speeds = pd.DataFrame({
    'nrows': nrows,
    'cpu_speeds': cpu_speeds,
    'gpu_speeds': gpu_speeds
})

speeds['difference'] = speeds['cpu_speeds'] - speeds['gpu_speeds']