We Keep Coding

Append std,mean columns to a DataFrame with a for-loop

I want to put the std and mean of a specific column of a dataframe for different days in a new dataframe. (The data comes from analyses conducted on big data in multiple excel files.)
I use a for-loop and append(), but it returns the last ones, not the whole.
here is my code:
hh = ['01:00','02:00','03:00','04:00','05:00']
for j in hh:
month = 1
hour = j
data = get_data(month, hour) ## it works correctly, reads individual Excel spreadsheet
data = pd.DataFrame(data,columns=['Flowday','Interval','Demand','Losses (MWh)','Total Load (MWh)'])
s_td = data.iloc[:,4].std()
meean = data.iloc[:,4].mean()
final = pd.DataFrame(columns=['Month','Hour','standard deviation','average'])
final.append({'Month':j ,'Hour':j,'standard deviation':s_td,'average':meean},ignore_index=True)

I am not sure, but I believe you should assign the final.append(... to a variable:
final = final.append({'Month':j ,'Hour':j,'standard deviation':x,'average':y},ignore_index=True)
Update
If time efficiency is of interest to you, it is suggested to use a list of your desired values ({'Month':j ,'Hour':j,'standard deviation':x,'average':y}), and assign this list to the dataframe. It is said it has better performance.(Thanks to #stefan_aus_hannover)

This is what I am referring to in the comments on Amirhossein's answer:
hh=['01:00','02:00','03:00','04:00','05:00']
lister = []
final = pd.DataFrame(columns=['Month','Hour','standard deviation','average'])
for j in hh:``
month=1
hour = j
data = get_data(month, hour) ## it works correctly
data=pd.DataFrame(data,columns=['Flowday','Interval','Demand','Losses (MWh)','Total Load (MWh)'])
s_td=data.iloc[:,4].std()
meean=data.iloc[:,4].mean()
lister.append({'Month':j ,'Hour':j,'standard deviation':s_td,'average':meean})
final = final.append(pd.DataFrame(lister),ignore_index=True)

Conceptually you're just doing aggregate by hour, with the two functions std, mean; then appending that to your result dataframe. Something like the following; I'll revise it if you give us reproducible input data. Note the .agg/.aggregate() function accepts a dict of {'result_col': aggregating_function} and allows you to pass multiple aggregating functions, and directly name their result column, so no need to declare temporaries. If you only care about aggregating column 4 ('Total Load (MWh)'), then no need to read in columns 0..3.
for hour in hh:
# Read in columns-of-interest from individual Excel sheet for this month and day...
data = get_data(1, hour)
data = pd.DataFrame(data,columns=['Flowday','Interval','Demand','Losses (MWh)','Total Load (MWh)'])
# Compute corresponding row of the aggregate...
dat_hh_aggregate = pd.DataFrame({['Month':whatever ,'Hour':hour]})
dat_hh_aggregate = dat_hh_aggregate.append(data.agg({'standard deviation':pd.Series.std, 'average':pd.Series.mean)})
final = final.append(dat_hh_aggregate, ignore_index=True)
Notes:
pd.read_excel usecols=['Flowday','Interval',...] allows you to avoid reading in columns that you aren't interested in the first place. You haven't supplied reproducible code for get_data(), but you should parameterize it so you can pass the list of columns-of-interest. But you seem to only want to aggregate column 4 ('Total Load (MWh)') anyway.
There's no need to store separate local variables s_td, meean, just directly use .aggregate()
There's no need to have both lister and final. Just have one results dataframe final, and append to it, ignoring the index. (If you get issues with that, post updated code here, make sure it's reproducible)

parsing CSV to pandas dataframes (one-to-many unmunge)

I have a csv file imported to a pandas dataframe. It probably came from a database export that combined a one-to-many parent and detail table. The format of the csv file is as follows:
header1, header2, header3, header4, header5, header6
sample1, property1,,,average1,average2
,,detail1,detail2,,
,,detail1,detail2,,
,,detail1,detail2,,
sample2, ...
,,detail1,detail2,,
,,detail1,detail2,,
...
(i.e. line 0 is the header, line 1 is record 1, lines 2 through n are details, line n+1 is record 2 and so on...)
What is the best way to extricate (renormalize?) the details into separate DataFrames that can be referenced using values in the sample# records? The number of each subset of details are different for each sample.
I can use:
samplelist = df.header2[pd.notnull(df.header2)]
to get the starting index of each sample so that I can grab samplelist.index[0] to samplelist.index[1] and put it in a smaller dataframe. Detail records by themselves have no reference to which sample they came from, so that has to be inferred from the order of the csv file (notice that there is no intersection of filled/empty fields in my example).
Should I make a list of dataframes, a dict of dataframes, or a panel of dataframes?
Can I somehow create variables from the sample1 record fields and somehow attach them to each dataframe that has only detail records (like a collection of objects that have several scalar members and one dataframe each)?
Eventually I will create statistics on data from each detail record grouping and plot them against values in the sample records (e.g. sampletype, day or date, etc. vs. mystatistic). I will create intermediate Series to also be attached to the sample grouping like a kernel density estimation PDF or histogram.
Thanks.

You can use the fact that the first column seems to be empty unless it's a new sample record to .fillna(method='ffill') and then .groupby('header1') to get all the separate groups. On these, you can calculate statistics right away or store as separate DataFrame. High level sketch as follows:
df.header1 = df.header1.fillna(method='ffill')
for sample, data in df.groupby('header1'):
print(sample) # access to sample name
data = ... # process sample records

The answer above got me going in the right direction. With further work, the following was used. It turns out I needed to use two columns as a compound key to uniquely identify samples.
df.header1 = df.header1.fillna(method='ffill')
df.header2 = df.header2.fillna(method='ffill')
grouped = df.groupby(['header1','header2'])
samplelist = []
dfParent = pd.DataFrame()
dfDetail = pd.DataFrame()
for sample, data in grouped:
samplelist.append(sample)
dfParent = dfParent.append(grouped.get_group(sample).head(n=1), ignore_index=True)
dfDetail = dfDetail.append(data[1:], ignore_index=True)
dfParent = dfParent.drop(['header3','header4',etc...]) # remove columns only used in
# detail records
dfDetail = dfDetail.drop(['header5','header6',etc...]) # remove columns only used once
# per sample
# Now details can be extracted by sample number in the sample list
# (e.g. the first 10 for sample 0)
samplenumber = 0
dfDetail[
(dfDetail['header1'] == samplelist[samplenumber][0]) &
(dfDetail['header2'] == samplelist[samplenumber][1])
].header3[:10]
Useful links were:
Pandas groupby and get_group
Pandas append to DataFrame

aggregating and pasting

I am beginning to move from R to Python and have a stupid question.
I have been looking for close to 5 hours to find a solution to my question.
I have the following code in R, which essentially takes the dataframe df and aggregates the outdates from a hospital based on unique ids. So my original table has many UIds repeated since someone may visit a hospital many times and each time they leave the hospital they have an out date. I want the UID, and all the outdates in one row. I could do this very easily with the following code in R.
newdf= aggregate(data = df, OutDate~UID, FUN=paste, sep="," )
Can anyone pray tell me how this can be accomplished in Python?
HEre's what my table looks like after using the above function in R
-UID1, 10/20/2008, 11/30/2008, 1/1/1900, 1/1/1900
-UID2, 6/19/2010, 1/1/1900
-UID3, 11/17/2009
-UID4, 3/14/2010 , 4/20/2010, 1/1/1900, 1/1/1900
-UID5, 12/12/2008, 8/27/2009, 1/1/1900
Ignore the dates, i just made them up. But the output needs to look like above.
Previously I had multiple UID1 rows for each of the dates in the current columns.
Now how do I do this in python.

You can do this with a dictionary comprehension:
from collections import defauldict
d = defaultdict(list)
for f in df.values():
// Assuming the first value is the UID:
d[f[0]].append(f)
Now d is a dictionary, where each key is the UID and the values are a list of rows from the dataframe. You can combine them into a string (like what you are doing with paste), like this:
for uid,values in d.iteritems():
for value in values:
print('{},{}'.format(uid,','.join(value)))

This sounds like building a dictionary where the key is the UID and you append each outdate to the key as you loop through the data. This assumes that you are getting the data in the form of a csv file where3 each row of data is read by csv.DictReader. I make the assumption based on what you seem to show of the data file and the separators. As a result, each entry in the row (which can include in time, out time, diagnosis, etc) is keyed by the header row. I will alsao assume that you can tell how to read the data into csv processing. The quick code below shows how to generate the dictionary entries from the row once you have it in.
I show the final way the data will look followed by how it was derived.
data = {UID1:(out1, out2, out3), UID2:(out3, out4)}
data = {}
for d in datarow:
uid = d[UID]
if uid not in data.keys():
data[uid] = ()
out = d[OUT]
data[uid].append(out)

Dealing with subsets of data using csv.DictReader

I'm parsing a big CSV file using csv.DictReader.
quotes=open( "file.csv", "rb" )
csvReader= csv.DictReader( quotes )
Then for each row I'm converting the time value in the CSV in datetime using this :
for data in csvReader:
year = int(data["Date"].split("-")[2])
month = strptime(data["Date"].split("-")[1],'%b').tm_mon
day = int(data["Date"].split("-")[0])
hour = int(data["Time"].split(":")[0])
minute = int(data["Time"].split(":")[1])
bars = datetime.datetime(year,month,day,hour,minute)
Now I would like to perform actions only on the rows of the same day. Would it be possible to do it in the same for loop or should I maybe save the data out per day and then perform actions? What would be an efficient way of baking the parsing?

As jogojapan has pointed out, it is important to know whether we can assume that the CSV file is sorted by date. If it is, then you could use itertools.groupby to simplify your code. For example, the for loop in this code iterates over the data one day at time:
import csv
import datetime
import itertools
with open("file.csv", "rb") as quotes:
csvReader = csv.DictReader(quotes)
lmb = lambda d: datetime.datetime.strptime(d["Date"], "%d-%b-%Y").date()
for k, g in itertools.groupby(csvReader, key = lmb):
# do stuff per day
counts = (int(data["Count"]) for data in g)
print "On {0} the total count was {1}".format(k, sum(counts))
I created a test "file.csv" containing the following data:
Date,Time,Count
1-Apr-2012,13:23,10
2-Apr-2012,10:57,5
2-Apr-2012,11:38,23
2-Apr-2012,15:10,1
3-Apr-2012,17:47,123
3-Apr-2012,18:21,8
and when I ran the above code I got the following results:
On 2012-04-01 the total count was 10
On 2012-04-02 the total count was 29
On 2012-04-03 the total count was 131
But remember that this will only work if the data in "file.csv" is sorted by date.

If (for some reason) you can assume that the input rows are already sorted by date, you could put them into a local container one by one as long as the date of any new row is the same as the previous one:
same_date_rows = []
prev_date = None
for data in csvReader:
# ... your existing code
bars = datetime.datetime(year,month,day,hour,minute)
if bars == prev_date:
same_date_rows.append(data)
else:
# New date. We process all rows collected so far
do_something(same_date_rows)
# Then we start a new collection for the new date
same_date_rows = [date]
# Remember the date of the current row
prev_date = bars
# Finally, process the final group of rows
do_something(same_date_rows)
But if you cannot make that assumption, you will have to
Either: Put the rows in a long list, sort that by date, and then apply an algorithm like the above to the sorted list
Or: Put the rows in a dictionary, using the date as key, and a list of rows as value for each key. Then you can iterate through the keys of that dictionary to get access to all rows that share a date.
The second of these two approaches is a little more space-consuming, but it may allow you do to some of the date-specific processing in the main loop, because whenever you receive a new row for an already-existing date, you could apply some of the date-specific processing right away, possibly avoiding the need to actually store all date-specific rows explicitly. Whether that is possible depends on what kind of processing you apply to the rows.

If you are not going for space efficeny, an elegant solution would be to create a dictionary where the key is your day, and the value is a list object, where all the information for each day is stored. Later you can do whatever operations you want based on per day.
For example
d = {} #Initialize emptry dictionry
for data in csvReader:
Day = int(data["Date"].split("-")[0])
try:
d[Day].append('Some_Val')
except KeyError:
d[Day] = ['Some_val']
This will either modify or create a new list object for each day. This is later easily accessible either by iterating over the dictionary or simply referring to the day as a key.
For example:
d[Some_Day]
will give you simply a list object with all the information you have stored. Given the linear lookup time of a dictionary, it should be quite efficent in terms of time.

Python in memory table data structures for analysis (dict, list, combo)

I'm trying to simulate some code that I have working with SQL but using all Python instead..
With some help here
CSV to Python Dictionary with all column names?
I now can read my zipped-csv file into a dict Only one line though, the last one. (how do I get a sample of lines or the whole data file?)
I am hoping to have a memory resident table that I can manipulate much like sql when I'm done eg Clean data by matching bad data to to another table with bad data and correct entries.. then sum by type average by time period and the like.. The total data file is about 500,000 rows.. I'm not fussed about getting all in memory but want to solve the general case as best I can,, again so I know what can be done without resorting to SQL
import csv, sys, zipfile
sys.argv[0] = "/home/tom/Documents/REdata/AllListing1RES.zip"
zip_file = zipfile.ZipFile(sys.argv[0])
items_file = zip_file.open('AllListing1RES.txt', 'rU')
for row in csv.DictReader(items_file, dialect='excel', delimiter='\t'):
pass
# Then is my result is
>>> for key in row:
print 'key=%s, value=%s' % (key, row[key])
key=YEAR_BUILT_DESC, value=EXIST
key=SUBDIVISION, value=KNOLLWOOD
key=DOM, value=2
key=STREET_NAME, value=ORLEANS RD
key=BEDROOMS, value=3
key=SOLD_PRICE, value=
key=PROP_TYPE, value=SFR
key=BATHS_FULL, value=2
key=PENDING_DATE, value=
key=STREET_NUM, value=3828
key=SOLD_DATE, value=
key=LIST_PRICE, value=324900
key=AREA, value=200
key=STATUS_DATE, value=3/3/2011 11:54:56 PM
key=STATUS, value=A
key=BATHS_HALF, value=0
key=YEAR_BUILT, value=1968
key=ZIP, value=35243
key=COUNTY, value=JEFF
key=MLS_ACCT, value=492859
key=CITY, value=MOUNTAIN BROOK
key=OWNER_NAME, value=SPARKS
key=LIST_DATE, value=3/3/2011
key=DATE_MODIFIED, value=3/4/2011 12:04:11 AM
key=PARCEL_ID, value=28-15-3-009-001.0000
key=ACREAGE, value=0
key=WITHDRAWN_DATE, value=
>>>
I think I'm barking up a few wrong trees here...
One is that I only have 1 line of my about 500,000 line data file..
Two is it seems that the dict may not be the right structure since I don't think I can just load all 500,000 lines and do various operations on them. Like..Sum by group and date..
plus it seems that duplicate keys may cause problems ie the non unique descriptors like county and subdivision.
I also don't know how to read a specific small subset of line into memory (like 10 or 100 to test with, before loading all (which I also don't get..) I have read over the Python docs and several reference books but it just is not clicking yet..
It seems that most of the answers I can find all suggest using various SQL solutions for this sort of thing,, but I am anxious to learn the basics of achieving the similar results with Python. As in some cases I think it will be easier and faster as well as expand my tool set. But I'm having a hard time finding relevant examples.
one answer that hints at what I'm getting at is:
Once the reading is done right, DictReader should work for getting rows as dictionaries, a typical row-oriented structure. Oddly enough, this isn't normally the efficient way to handle queries like yours; having only column lists makes searches a lot easier. Row orientation means you have to redo some lookup work for every row. Things like date matching requires data that is certainly not present in a CSV, like how dates are represented and which columns are dates.
An example of getting a column-oriented data structure (however, involving loading the whole file):
import csv
allrows=list(csv.reader(open('test.csv')))
# Extract the first row as keys for a columns dictionary
columns=dict([(x[0],x[1:]) for x in zip(*allrows)])
The intermediate steps of going to list and storing in a variable aren't necessary.
The key is using zip (or its cousin itertools.izip) to transpose the table.
Then extracting column two from all rows with a certain criterion in column one:
matchingrows=[rownum for (rownum,value) in enumerate(columns['one']) if value>2]
print map(columns['two'].__getitem__, matchingrows)
When you do know the type of a column, it may make sense to parse it, using appropriate
functions like datetime.datetime.strptime.
via Yann Vernier
Surely there is some good reference for this general topic?

You can only read one line at a time from the csv reader, but you can store them all in memory quite easily:
rows = []
for row in csv.DictReader(items_file, dialect='excel', delimiter='\t'):
rows.append(row)
# rows[0]
{'keyA': 13, 'keyB': 'dataB' ... }
# rows[1]
{'keyA': 5, 'keyB': 'dataB' ... }
Then, to do aggregations and calculations:
sum(row['keyA'] for row in rows)
You may want to transform the data before it goes into rows, or use a friendlier data structure. Iterating over 500,000 rows for each calculation could become quite inefficient.
As a commenter mentioned, using an in-memory database could be really beneficial to you. another question asks exactly how to transfer csv data into a sqlite database.
import csv
import sqlite3
conn = sqlite3.connect(":memory:")
c = conn.cursor()
c.execute("create table t (col1 text, col2 float);")
# csv.DictReader uses the first line in the file as column headings by default
dr = csv.DictReader(open('data.csv', delimiter=','))
to_db = [(i['col1'], i['col2']) for i in dr]
c.executemany("insert into t (col1, col2) values (?, ?);", to_db)

You say """I now can read my zipped-csv file into a dict Only one line though, the last one. (how do I get a sample of lines or the whole data file?)"""
Your code does this:
for row in csv.DictReader(items_file, dialect='excel', delimiter='\t'):
pass
I can't imagine why you wrote that, but the effect is to read the whole input file row by row, ignoring each row (pass means "do exactly nothing"). The end result is that row refers to the last row (unless of course the file is empty).
To "get" the whole file, change pass to do_something_useful_with(row).
If you want to read the whole file into memory, simply do this:
rows = list(csv.DictReader(.....))
To get a sample, e.g. every Nth row (N > 0), starting at the Mth row (0 <= M < N), do something like this:
for row_index, row in enumerate(csv.DictReader(.....)):
if row_index % N != M: continue
do_something_useful_with(row)
By the way, you don't need dialect='excel'; that's the default.

Numpy (numerical python) is the best tool for operating on, comparing etc arrays, and your table is basically a 2d array.