</div>

</div>



# Pandas (like the rest of Python) is object-oriented

- Pandas is object-oriented.

- We create data frames by constructing instances of different classes.

- The two most important classes are:

    - `DataFrame`
    - `Series`
    
- Pandas follows the Java convention of starting the name of classes with an upper-case letter, whereas instances are all lower-case.




## Pandas Series
- One-dimensional array 
- Series can be like array, with standard integer index starting at 0
- Series can be dictionary like, with defined index 




<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```data = [1,2,3,4,5] #This creates a list
my_series = pd.Series(data) #array-like pandas series, index created automatically
my_series2 = pd.Series(data, index=['a', 'b', 'c', 'd', 'e'])  #dict like, index specified
print(my_series, '\n', my_series2)

</div>

</div>



# Creating a Series from a `dict`





<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```d = {'a' : 0., 'b' : 1., 'c' : 2.}
my_series = pd.Series(d)
my_series

</div>

</div>



## Functions on Series

- We can perform calculations using the entire series similar to numpy.
- Methods are called from within `np.Series`, for example np.Series.add
- See a variety of series functions [here](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.Series.html) 





<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#These are just a variety of examples of operations on series.


starter = {'a' : 0., 'b' : 1., 'c' : 2.}

a = pd.Series(starter)
print('Print the entire array a: \n', a)

b1=10*a
print('Mulitiply by 10 directly:\n', b1)

b2=a.multiply(10)
print('Mulitiply by 10 using the function:\n', b2)

c1=a+b1
print('Add a and b together directly:\n', c1)

c2=pd.Series.add(a,b1) #Note we are calling the method of the series class. Numpy used us np.add
print('Add a and b together with a function:\n', c2)


suma=pd.Series.sum(a) #Note we are calling the method of the series class. Numpy used us np.add
print('sum all of a:\n', suma)


f=a**2  #This squares the value. 
print('square a:\n', f)
x = pd.Series({'a' : 0., 'b' : 1., 'c' : 2.})
y = pd.Series({'a' : 3., 'b' : 4., 'c' : 5.})
z = x+y
print('Add 2 series together:\n', z)

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```time_series = pd.Series(data, index=dates)
time_series

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```type(time_series)

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```type(d)

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```df

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```df2=df ** 2 #This squares all values. 
print(df2)

</div>

</div>



### Data Types
- Each will have an inferred data type. 



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```print(df.dtypes)

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```# Pulling from a local file
frame2 = pd.read_csv('iris.csv')
frame2

</div>

</div>



## Indexing/Slicing Rows of DataFrames
- Simple ways of selecting all rows and colu (`df[:]`)
- Rows can be accessed via a key or a integer corresponding to the row number. 
- Omitting a value generally means *all values* before or after an item.
- When we retrieve a single or mulitiple rows, the result is a Dataframe.
- Several ways, either directly, with `iloc`, or with `loc`. (See Examples).
- Read more [here](http://pandas.pydata.org/pandas-docs/stable/indexing.html)





<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#This is going to create some sample data that we can work with for our analysis. 

import pandas as pd
import numpy as np
 
#Create a dataframe from a random numpy array 
#http://docs.scipy.org/doc/numpy/reference/generated/numpy.random.randn.html
df = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['a', 'b', 'c', 'd'] )
print (df)
df2 = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['e', 'f', 'g', 'h'] )
print (df2)

</div>

</div>



## Dropping Columns from Dataframes
- Done using the `drop` syntax. 
- [Drop Documentation](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.drop.html)



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#Here, we can remove columns specifically from a dataframe using the drop method.
df2 = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['e', 'f', 'g', 'h'] )
print (df2)
df2.drop(['e','f'], inplace=True, axis=1)
print (df2)

## Selecting rows
print (df[0:3])     #Select rows 1-3
print (df.iloc[0:3,:])     #Select rows 1-3
print (df.iloc[0:3,])      #Select rows 1-3
print (df.iloc[0:3])       #Select rows 1-3
print (df.iloc[[1,2,4]])   #Select rows 1, 2, and 4

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#notice how the logical statement is inside the dataframe specification.  This creates an intermediate boolean array. 
df[df['a'] >= 20]

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#We can now generate a vector based on a critera and then use this for selection
select = df['a']>=20
print (select,type(select))
print (df.loc[select,'a']) #Notice by including only one variable we are selecting rows and all columns.

select2 = (df['a']>20) & (df['c'] < 30)  #More complex criteria
print (select2)
print (df.loc[select2,['a','c']])

</div>

</div>



## Joining Dataframes
- Often you need to combine dataframe, 
- either matching columns for the smae rows (column  bind)
- Add rows for the same columns (row bind)



## Stacking Dataframes Vertically
- Adds rows vertially with the `concat` function
- The index is not automatically reset
- In R referred to as a row bind.



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#This first generates 2 dataframes. 
df = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['a', 'b', 'c', 'd'] )
df2 = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['a', 'b', 'c', 'd'] )

#This will stack the 2 dataframes vertically on top of one another
dfbyrow=pd.concat([df, df2])  #This is equivalent to a rowbind in R. 
print (dfbyrow)

</div>

</div>



## Inner/Outer Joins Dataframes
- Adds rows vertially with the `concat` function
- In R referred to as a column bind.
- Can do the equivalent of an inner and outer join.



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#Merging additional columns also uses the concat function 
#This is equavalent to an inner join in SQL.
df = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['a', 'b', 'c', 'd'] )
df4 = pd.DataFrame((20+np.random.randn(10, 4)*5),  columns=['e', 'f', 'g', 'h'] )


dfbycolumns = pd.concat([df, df4], axis=1, join='inner')
dfbycolumns

</div>

</div>



<div markdown="1" class="cell code_cell">
<div class="input_area" markdown="1">
```#Here, the outer does the equivalent of a left outer join for this dataset. 
shortdf=df[0:5]
dfbycolumns = pd.concat([df, shortdf], axis=1, join='outer')
dfbycolumns