Python Pandas Notebook

learningPandas.md

Things that I need to be able to do

Import from csv into memory

To create a dataframe from a csv, you can use the following:

df = pd.read_csv('pandas_dataframe_importing_csv/example.csv')

If you have a different delimiter than a comma, you can use the sep parameter:

df = pd.read_csv('pandas_dataframe_importing_csv/example.csv', sep="\|", engine="python")

Notice the escape character before the delimiter character.

With no headers, you need to add the following:

df = pd.read_csv('pandas_dataframe_importing_csv/example.csv', sep="\|", engine="python", header=None)

That will give you something like the following:

                          0            1    2       3
0  2018-06-22T23:59:47.965Z  123-456-789  200  12.203
1  2018-06-22T23:60:47.965Z  132-456-789  200  14.203
2  2018-06-22T23:61:47.965Z  312-456-789  400  15.203
3  2018-06-22T23:62:47.965Z  231-456-789  200  11.203

Calculate the min for a column

Assuming that you know the column you want to take the min of, you'll use the column number as an index to the dataframe. For example, just using the index gives you all the column values:

In [13]: df[3]
Out[13]:
0    12.203
1    14.203
2    15.203
3    11.203
Name: 3, dtype: float64

You can then take the df[i] format and tag it with the .min() method to give you the min value:

In [11]: df[3].min()
Out[11]: 11.203

Calculate the max for a column

Same as with the .min(), only we're switching out for the .max() method:

In [12]: df[3].max()
Out[12]: 15.203

Calculate the average for a column

Second grade fun fact: mean = average. Evidently I lost that knowladge over the years.

Same as min and max:

In [19]: df[3].mean()
Out[19]: 13.203

Fun side note... use describe()

You can use the describe() method a dataframe to get a whole bunch of info on all of your numeric columns super quickly. See the following:

In [18]: df.describe()
Out[18]:
           2          3
count    4.0   4.000000
mean   250.0  13.203000
std    100.0   1.825742
min    200.0  11.203000
25%    200.0  11.953000
50%    200.0  13.203000
75%    250.0  14.453000
max    400.0  15.203000

Calculate the 95th percentile of a column

You can calculate percentiles by using the quantile() method. This takes in some fractional value and returns that percentile. For example, to get the 95th, you'd use .95:

In [28]: df[3].quantile(.95)
Out[28]: 15.052999999999999

Calculate the 99th percentile of a column

Same as the previous, only using .99:

In [26]: df[3].quantile(.99)
Out[26]: 15.173

Filter by column value

You can filter down columns by doing conditional evaluations, and even combine them to do more complex queries:

In [40]: filtered_data = df[df[0] > '2018-06-22T23:60:47.965Z']

In [41]: filtered_data
Out[41]:
                          0            1    2       3
2  2018-06-22T23:61:47.965Z  312-456-789  400  15.203
3  2018-06-22T23:62:47.965Z  231-456-789  200  11.203

In [42]: filtered_data = df[(df[0] > '2018-06-22T23:60:47.965Z') & (df[2] != 200)]

In [43]: filtered_data
Out[43]:
                          0            1    2       3
2  2018-06-22T23:61:47.965Z  312-456-789  400  15.203

Turn a column into a list

Surprise, there's another built in method to handle this too. Using the tolist() method, you can turn a specific column into a list:

In [30]: df[3].tolist()
Out[30]: [12.203, 14.203, 15.203, 11.203]

Compare csv lengths (whatever that term is in pandas)

Another method, count(), is your friend:

In [32]: df[3].count()
Out[32]: 4

However, this is potentially dangerous as it will only count rows where non NaN values are present. In that case, you can use shape[0]:

n [36]: df.shape[0]
Out[36]: 4

Handling dates

When importing in the csv, you need to use the parse_dates property to set equal to the columns that have dates in them:

In [37]: df = pd.read_csv('t1.csv', sep="\|", engine="python", header=None, parse_dates=[0])

Merge two csv by value in column

Can be done using the merge method. The two tables will then create a new dataframe based on the key that you provided:

In [46]: df = pd.merge(df1, df2, on=[1], how='left', indicator='Exist')

In [47]: df
Out[47]:
                        0_x            1  2_x     3_x                       0_y  2_y     3_y Exist
0  2018-06-22T23:59:47.965Z  123-456-789  200  12.203  2018-06-23T23:59:47.965Z  200  12.303  both
1  2018-06-22T23:60:47.965Z  132-456-789  200  14.203  2018-06-23T23:60:47.965Z  200  14.303  both
2  2018-06-22T23:61:47.965Z  312-456-789  400  15.203  2018-06-23T23:61:47.965Z  400  15.303  both
3  2018-06-22T23:62:47.965Z  231-456-789  200  11.203  2018-06-23T23:62:47.965Z  200  11.303  both

If the key can't be found, then a NaN value will appear in the merged version:

In [49]: df = pd.merge(df1, df2, on=[1], how='left', indicator='Exist')
In [56]: df['Exist'] = np.where(df.Exist == 'both', True, False)

In [50]: df
Out[50]:
                        0_x            1  2_x     3_x                       0_y    2_y     3_y      Exist
0  2018-06-22T23:59:47.965Z  123-456-789  200  12.203  2018-06-23T23:59:47.965Z  200.0  12.303       True
1  2018-06-22T23:60:47.965Z  132-456-789  200  14.203  2018-06-23T23:60:47.965Z  200.0  14.303       True
2  2018-06-22T23:61:47.965Z  312-456-789  400  15.203  2018-06-23T23:61:47.965Z  400.0  15.303       True
3  2018-06-22T23:62:47.965Z  231-456-789  200  11.203                       NaN    NaN     NaN  False

Check to see if all values are present

Can either search on the number of NaNs that are present in the table, or off of the created Exists column that we've overwritten to be True or False if the record exists:

In [53]: df.isnull().sum().sum()
Out[53]: 3

In [54]: df.isnull().sum()
Out[54]:
0_x      0
1        0
2_x      0
3_x      0
0_y      1
2_y      1
3_y      1
Exist    0
dtype: int64

In [55]: df['0_y'].isnull().sum()
Out[55]: 1

In [58]: df[df['Exist'] == False]
Out[58]:
                        0_x            1  2_x     3_x  0_y  2_y  3_y  Exist
3  2018-06-22T23:62:47.965Z  231-456-789  200  11.203  NaN  NaN  NaN  False

11. Generate graphs