df.info(memory_usage="deep")  # memory_usage: 是否精确显示内存使用量

# 查看字段类型的表达范围
np.iinfo("int32")

# 选择相同类型的字段
cvt = df.select_dtypes(include=['int'])

# 多列转换为数值型字段
cvt.apply(pd.to_numeric, downcast="signed")  # downcast: signed / unsigned / float

# 离散字段存储方式
df['col'].astype('category') # 当 nunique / total <= 0.5 时采用

converted_obj = pd.DataFrame()

for col in gl_obj.columns:
    num_unique_values = len(gl_obj[col].unique())
    num_total_values = len(gl_obj[col])
    if num_unique_values / num_total_values < 0.5:
        converted_obj.loc[:,col] = gl_obj[col].astype('category')
    else:
        converted_obj.loc[:,col] = gl_obj[col]


# 在读取 csv 文件时，以字典的方式指定读取的字段类型
column_types = {
   'acquisition_info': 'category',
    'h_caught_stealing': 'float32',
    'h_player_1_name': 'category'
}
read_and_optimized = pd.read_csv('game_logs.csv',dtype=column_types,parse_dates=['date'],infer_datetime_format=True)