#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import pandas as pd
import numpy as np

## 讀取資料
df = pd.read_csv("./HR.csv", header=0)

# 檢視資料結構
summary = df.describe()

# 求均值
row_mean = df.mean(axis=1)
col_mean = df.mean()

# 選擇資料
## 列
print(df["satisfaction_level"].head())
print(df[0:3])

## 標籤
print(df.loc[0:3])
print(df.loc[0, ["satisfaction_level"]])


### 1
### 異常值分析
### 空值處理
sl_l = df["satisfaction_level"]
df[df['satisfaction_level'].isnull()]
#print(sl_l.isnull())
print(sl_l.isnull().sum())
print(sl_l[sl_l.isnull()])
## 對空值的填充
#print(sl_l.fillna(value=5))

## 對空值的丟棄
#print(sl_l.dropna(how="any"))
sl_l = sl_l.dropna(how="any")


### 2
### 資料過大、過小異常處理
le_s = df['last_evaluation']
le_s[le_s.isnull()]
le_s.isnull().sum()

## 偏度
le_s.skew()
## 峰度
le_s.kurt()

## 連續異常值處理方式(取四分位上下界)
#（1） le_s = le_s[le_s <= 1]
q_low = le_s.quantile(q=0.25)
q_high = le_s.quantile(q=0.75)
q_interval = q_high - q_low
k = 1.5

### 資料篩選
le_s = le_s[le_s<q_interval+k*q_interval][le_s>q_low-k*q_interval]

### 分佈情況
np.histogram(le_s.values,bins=np.arange(0.0, 1.1,0.1 ))

### 3
## 排序
np_s = df['number_project']
np_s.value_counts(normalize=True).sort_index()

### 4 分佈情況
pl5_s = df['promotion_last_5years']
pl5_s.value_counts()
pl5_s.value_counts(normalize=True)

## 5 條件篩選
s_s = df['salary']
s_s.where(s_s!=="nme").dropna()


### 總結 
# 去空值
df = pd.read_csv("./HR.csv", header=0)
df = df.dropna(axis=0,how='any')

df[df['last_evaluation']<=1][df['salary']!='nme']
le_s = df['last_evaluation']
q_low = le_s.quantile(q=0.25)
q_high = le_s.quantile(q=0.75)
q_interval = q_high - q_low
k=1.5
le_s = le_s[le_s<k*q_interval+q_high][le_s>k*q_interval-q_low]
df[le_s<k*q_interval+q_high][le_s>k*q_interval-q_low][df['salary']!='nme']
簡單對比分析
df.groupby("department").mean()
 

待續。。。