Wednesday, October 1, 2014

My Car Buying Learnings

I recently went through a car buying ordeal and wanted to share some points that worked for me. Some of them were things that I learnt from reviewing the plethora of car buying advice on the internet and some of them were pure luck/thinking on the feet.

Which car to buy is another challenging decision to make. The points below make sense after you have narrowed down the car that you want to buy. 

1. Buy Consumer Reports report on your vehicle(s). This is an under $20 investment for a potential few thousand dollars purchase. So totally worth it. 

2. Get Pre-Approved on a loan, preferably from a Federal Credit Union (because they offer the lowest rates, per my research, at this time).

3. Ask for email quotes from multiple dealers. This way you can avoid visiting multiple dealerships and wasting time.

4. Focus on the out the door $$$. Ignore "free" throw ins like, tire locks, 2 year free maintenance etc. Recruit a buddy to constantly enforce this. I kept getting tempted to go down the 'throw-ins' path. Fortunately, my mother was my buddy who kept me from chasing the shiny object.

5. Pick a month when the month end is on Monday or a Sunday. In hind sight, I believe that this double whammy really worked for me.

6. Reach the dealer to sign the deal on Sunday late afternoon, as they are in a hurry to pack up as well. 4 PM is ideal.

7. Be ready to walk away, if anything does not feel right.

8. Follow the Consumer Reports Car Buying Guide to the 'T'.

9. Negotiate on the Financing Rate based on #2 above.

10. Negotiate on the Extended Warranty $$$ if you are getting one. Pay it outside the loan amount for lesser finance charge. I did not do this and in hind sight, should have.

11. Ask to waive the Doc fees.

12. Pitch one dealer against the other, ask to provide a better offer. Do not feel bad about this. Have a buddy constantly remind you to not feel bad about this.

13. One learning that I found missing in all the research I did was that Buyers Remorse is for real and you should prepare for it. If you are, like me, financially conservative (my wife prefers the term "stingy"), then this will hit you after the deal is closed. Acknowledging it makes it easier to manage it.

One litmus test, I discovered, to measure if you did good is how annoyed was the sales guy when you were signing the deal. The value of the deal is directly proportional to the  pissed off factor!

Do you have any tips or experiences to share on car buying? How about on home buying? Feel free to share those in the comments.

Tuesday, September 16, 2014

Pycasa - Group duplicate files

A problem that I have been battling is how unorganized my photos and videos are. I have multiple backups taken from the various devices, which has resulted in multiple copies of images and videos. Recently, I started thinking about how to organize those with the following goals:


  • Remove duplicates
  • Organize by year and by event
  • Remove images that are out of focus or not relevant


After manually going through a few folders, I realized that it was going to be a very onerous task. So I started thinking about writing a program to identify duplicates. I recently watched a great video by David Beazley on Python generators and it was exactly what I needed for my purposes.

So I created this program that groups the duplicates together and prints a list for review.

import os
import time
import fnmatch
from collections import defaultdict

#http://szudzik.com/ElegantPairing.pdf
#This is used to combine size_of_file_in_bytes and last_modified into one number associated uniquely to both
#using a pairing function.
def elegant_pairing_fn(x,y):
    if x == max(x,y):
        return (x*x)+x+y
    else:
        return (y*y)+x

#generates a sequence of file names for the given search pattern starting from the top directory.
def gen_find(filepat, top):
    for path, dirlist, filelist in os.walk(top):
        for name in fnmatch.filter(filelist, filepat):
            yield os.path.join(path,name)

#generates a sequence of dictionary objects with detailed file information.
def gen_stat(filelist):
    for name in filelist:
        dstat = (dict(zip(osstatcolnames,os.stat(name))))
        dstat['filename'] = name
        dstat['unique_id'] = elegant_pairing_fn(dstat['size_of_file_in_bytes'], dstat['last_modified'])
        yield dstat

osstatcolnames=('protection_bits', 'inode_number', 'device', 'number_of_hard_links'
          , 'user_id_of_owner', 'group_id_of_owner', 'size_of_file_in_bytes'
          , 'last_accessed', 'last_modified', 'created')

t0 = time.time()
jpgfiles = gen_find("*.jpg", "C:\\Users\\Shantanu\\Pictures")
jpgfilestats=gen_stat(jpgfiles)

#Now group the file names with the same unique_id value
#The algorithm assumes that the combination of file size and last modified date is unique per image file.
# Meaning it is almost impossible to have images that are different and yet have the same size and last modified date.
# This is a safe assumption as long as the files have not been modified using some photo editing software.
filecount = 0
jpgfilegroups = defaultdict(list)
for l in jpgfilestats:
    filecount += 1
    jpgfilegroups[l['unique_id']].append(l['filename'])

#Print groups with more than one file names in it.
#  We do not care to review files that appear only once in our stash.
count = 0
groupcount = 0
for k,v in jpgfilegroups.items():
    groupcount += 1
    if len(v) > 1:
        count += 1
        print(count, k, v)
print ("Total files processed: ", filecount)
print ("Total file groups processed: ", groupcount)
print ("Total time taken: ", time.time() - t0)


On my laptop, the program processed ~11K files in 2.8 seconds and came up with 2689 groups of possible duplicates. Manually reviewing those 2689 groups is a much simpler task.

The algorithm assumes that the combination of file size and last modified date is unique per image file. Meaning it is almost impossible to have images that are different and yet have the same size and last modified date. This is a safe assumption as long as the files have never been modified using some photo editing software (which is conveniently true for my case).

I learnt some new tools as a result of this exercise:

  • A practical use of Python generators for scratching my own itch.
  • The Python os library and its functions
  • The mathematical concept of 'pairing functions' to uniquely represent two numbers as one.
  • EDIT 20141020: I just realized that jpgfilegroups is actually a hash table using chaining for collision resolution.
What problems have you solved using these tools? Feel free to use this code to organize your photos and videos.

Monday, September 15, 2014

My Latest Laptop

I wrote the first half of this post in 2011 with an update in February 2014.

After 7 years of superb performance from my IBM Thinkpad T40, it finally died of a power connection break inside the Motherboard. After doing much research, I decided to stick to Lenovo instead of changing brands.

I noticed in my behavior and of some of my friends too, that one doesn't switch laptop brands that easily.

So here is what I ended up with:
Lenovo Z570
Second Generation i7 (2.0 GHz)
8 GB RAM
240 GB OCZ Vertex 3 SATA 3 6GBPS SSD
700 GB 5400 RPM WD HDD
External Optical Drive

Here is how much it cost me (including taxes & shipping):
1. $10.59 - For 3 new Philip Head Screwdrivers
2. $32.85 - SATA - ESata connector cable for External Optical Drive
3. $51.65 - Optical Drive HDD Caddy
4. $742.69 - Lenovo Z570 Laptop
5. $49.99 - Acronis True Home Image
6. $459.99 - OCZ Vertex 3 SSD 250 GB
For a total of - $1347.76

That's the price I paid for a laptop with:
1. Almost 1 TB of total disk space
2. Super Fast SSD Performance (~20 Second Boot times)
3. Security of backups given the instability of SSDs!
4. External Optical Drive that is not taking up space in the laptop.

Update 20140205: Return of the IBM Thinkpad T40.
First, an update on the Z570. It has been working great. Like Jeff Atwood of Coding Horror fame says "A solid state hard drive is easily the best and most obvious performance upgrade you can make on any computer for a given amount of money. Unless your computer is absolute crap to start with.". It is well worth the price. Also, I have been fortunate that I have not had any catastrophic SSD failures yet.

Going back to the T40. I am glad that I did not throw it away. I was able to move most of the working parts from this T40 to another one that the IT department at work was trashing. I moved the HDD, Internal Wireless Card, Internal Bluetooth Card, RAM to the new chassis.

In turn, I learnt a lot about the inside of a laptop. Most of it is like a jigsaw puzzle. The parts are built in such a way that they fit in perfect slots, mostly. The biggect challenge I had was keeping a track of all the screws. Everytime I opened the laptop, I ended up with a few screws that I could not figure out where I took them out from!

One change that I did make was to move to Lubuntu 12.04. With Windows XP end of life in April 2014, and knowing the fact that the T40 hardware is too weak for Windows 7 and above, I decided to switch. And that was a smart move.

By switching to Lubuntu, I have extended the life of my T40 by at least another 2 years.

Now all I need is a new battery pack, 1 GB of RAM and a 60 GB SSD (SATA 2 will be fine). I will then put the current HDD into the optical drive bay and make the SSD the master. I should end up with the meanest T40 out there!

Do you have any insights to share on ways to alter laptops to make them more useful from a practical point of view? 

Saturday, August 9, 2014

Python GroupBy, Map & Reduce

I came across a really interesting data mangling technique while watching this presentation on advanced Python programming techniques

Here is the example from this talk. Suppose you have a list of dictionary objects, sorted by the 'id' key, my_list as defined below.
>>> my_list = [
    {'id':1, 'name':'raymond'},
    {'id':1, 'email':'ray@spkrbar.com'},
    {'id':2, 'name':'sue'},
    {'id':2, 'email':'sue@sally.com'}] #sorted

Using dictgroupby, map and reduce, there is a very elegant way of grouping all of those dictionaries by the 'id' to get the following list:

[
  {'id': 1, 'email': 'ray@spkrbar.com', 'name': 'raymond'}, 
  {'id': 2, 'email': 'sue@sally.com', 'name': 'sue'}
]

And here is how you do it:
>>>from itertools import groupby
>>> [dict (
    reduce(lambda y,z: y + z,
        map(lambda x: x.items(), v)
    )
)
for k, v in groupby(my_list, key=lambda x: x['id'])]

Notice how much this is 'SQL' like. Let us break this statement down to its individual components to understand what is happening under the covers. Like any SQL statement, we have to start deciphering it inside out.

Let us look at the for loop with the groupby operation in it. The groupby operation will return an iterator grouping by the 'key' parameter. In this case, it is an anonymous function that returns the 'id' value. Essentially, we are asking for the grouping to happen using the 'id' values (1, 2 etc.).

>>>print({k:list(v) for k,v in groupby(my_list, key=lambda x: x['id'])})

{
 1: [{'id': 1, 'name': 'raymond'}, 
     {'id': 1, 'email': 'ray@spkrbar.com'}], 
 2: [{'id': 2, 'name': 'sue'}, 
     {'id': 2, 'email': 'sue@sally.com'}]
}

Note that the 'id' values are the keys and they are also repeated as part of the values. This will come in handy for the next step.

Next, we map the anonymous function, which calls the items() method for the parameter passed in, over each of the groups returned from the groupby operation.

>>> for k, v in groupby(my_list, key=lambda x: x['id']):
...     print(map(lambda x: x.items(), v))
...
[[('id', 1), ('name', 'raymond')], [('id', 1), ('email', 'ray@spkrbar.com')]]
[[('id', 2), ('name', 'sue')], [('id', 2), ('email', 'sue@sally.com')]]

This gives us lists of tuples instead of a dictionaries, which makes the reduction step very easy.

Now, we reduce the list that comes out of the mapping step by doing a simple addition of lists. Addition over two lists results in a list with elements from both. We would not have been able to use the '+' operator if these were dictionaries instead. Notice also the duplicate 'id' tuple.
>>>print(reduce(lambda y,z: y + z, [[('id', 1), ('name', 'raymond')], [('id', 1), ('email', 'ray@spkrbar.com')]]))
[('id', 1), ('name', 'raymond'), ('id', 1), ('email', 'ray@spkrbar.com')]

The reduce step will also happen for the list for 'id' 2 in this example.

We are almost at the end now. The last step is to make a dictionary from the list coming out of the reduce step to remove duplicates and conform back to the input which was a list of dictionaries.

>>>print(dict([('id', 1), ('name', 'raymond'), ('id', 1), ('email', 'ray@spkrbar.com')]))
{'id': 1, 'email': 'ray@spkrbar.com', 'name': 'raymond'}

Note that the duplicate 'id' tuple got removed. As before, this step will also happen for the list for 'id' 2.

That is it! Now, we have what we need. As list of dictionary objects, grouped by 'id'.

[
  {'id': 1, 'email': 'ray@spkrbar.com', 'name': 'raymond'}, 
  {'id': 2, 'email': 'sue@sally.com', 'name': 'sue'}
]

What are some of the practical uses you see for this technique? Do you have any other slick trick to share?Let me know in the comments below.

Friday, September 14, 2012

How a Sub Registrar Officer does corruption

I have been following the India Against Corruption movement for the past 6 months. This movement inspired me to blog about my experience with the Sub Registrar Office in Pune, India where I witnessed corruption happening in the first hand.

The Right To Information act has made it a little difficult for corruption to go unnoticed. However, one of the biggest contributor to corruption is public ignorance. Let me spell out the details of how the events unfolded and explain this point.

I was trying to give my Father the Power Of Attorney so that he could do transactions on my behalf in Pune. In order to do that, one has to type up a 5-6 page document that states, in legal jargon, that you are giving the other person the right to act on your behalf. There are many templates floating around that are used for this. Since one of my distant relations happens to have a few lawyers on his staff, we approached them for some legal advice. Naturally, they got us the template, we filled in the blanks, and now we had a document ready.

This document really doesn't mean a lot until it has been registered by the State Department of Registration and Stamps.

Now, we knew that we had to get this document registered. So we asked the lawyers at our disposal how to proceed. And this is where our ignorance got the best of us. The lawyers immediately referred us to their "agent" who "knows" the process and will "get things done" for us. Obviously for a nominal fee!

The Department of Registration and Stamps has a website and there they have instructions on what is required to get your documents registered. If the lawyers, would have pointed us to this link, we would have been saved.

Anyway, we contacted the agent and asked him what was a good time to meet him and setup the "terms" of our transaction. It was decided that we should come to the Sub Registrar Office where he "works" the next day and he will have our job done within a few hours. As the ignorant fools we were, we fell for it.

So, this is how it works. This "agent" is sometimes a lawyer by education. We are not talking about the cream of the crop here, definitely. This guy has managed to get the LLB degree and is using a little miscommunication and a little bribery to get an "edge". So what is the "edge" you ask? Let me elaborate.

The department website allows you to make an online appointment for your business at the office. Once you have a token, you are expected to appear, with all documents and personnel, 30 minutes before your time of appointment and the rest will take care of itself. The Sub Registrar Officer, however, has the authority to book time-slots from his terminal. And this is how he can provide an "edge" to the "agents". For some money in cash, he hands out time-slots to the "agents". The "agents" now sell those time-slots to their customers by telling them when they should come to the office and be gone once complete. The reason the "agent" has an LLB degree is that you can substitute one lawyer in lieu of two witnesses for the registration. Nowhere on the website does it say that it is okay to bring in one lawyer instead of two witnesses. This "one lawyer" clause is written on a board in the Sub Registrar Office.

So, by selling appointments, that are available online for free, the Sub Registrar Officer is able to make a few bucks on the side.

It is not often that people go to the Sub Registrar Office. So, the attitude they have is that might as well pay a few extra rupees and be done with it. However, from my experience, I did not get any "edge" and it would have been much better if I would have used the website.

The lesson(s) I learnt from this experience is/are:
  • Check the government websites on what is needed for your work.
  • Don't follow the crowd by approaching an "agent".
  • Right To Information Act is a very powerful tool, use it to your advantage
What experiences have you had that would allow other's to not fall into such ignorance traps?

Tuesday, May 8, 2012

Umass Boston CS Alumni Speech 2012

I was invited as the speaker at the 2012 Alumni Party held by the CS department at UMass Boston.
It was an honor and a privilege to speak to the wonderful audience.
Here is the presentation.

Wednesday, March 3, 2010

Peter Norvig's Spelling Corrector in VB

Some Background: The idea behind this implementation was not to build the shortest or the fastest version of the spelling corrector. I looked at the list of languages that this was implemented in and found VB .NET missing. So I decided to fill that void. I also thought of it as a way to unite the cult of VB .NET programmers with the others.
(!!!Noble Peace Prize nomination here please!!!)

More importantly, I wanted to spell out each of the steps to make it easier to understand the concept. I also wanted to use native libraries so that you can dive into the spelling corrector concepts quickly without first having to learn other technologies like LINQ etc.

Please feel free to post any comments, suggestions for improvement and any bugs you find.

Copy the text below into a VB class file and get the BIG.txt file (http://norvig.com/big.txt).

' VB .NET Implementation of Peter Norvig's Spelling Corrector.
' VERSION 1.0, last updated 03 Mar 10.
'
' Peter Norvig's original article located at
' http://norvig.com/spell-correct.html
'
' Copyright (c) 2010 Shantanu Inamdar
'
' Permission is hereby granted, free of charge, to any person obtaining a copy
' of this software and associated documentation files (the "Software"), to deal
' in the Software without restriction, including without limitation the rights
' to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
' copies of the Software, and to permit persons to whom the Software is
' furnished to do so, subject to the following conditions:

' The above copyright notice and this permission notice shall be included in
' all copies or substantial portions of the Software.

' THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
' EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
' MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
' IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR
' ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
' CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
' WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.


' Some Background: The idea behind this implementation was not build the shortest
' or the fastest version of the spelling corrector.
' I looked at the list of languages that this was implemented in and found VB .NET
' missing. So I decided to fill that void. I also thought of it as a way to unite
' the VB .NET programmers with the others.
' (!!!Noble Peace Prize nomination here please!!!)
'
' More importantly, I wanted to spell out each of the steps to make it easier to
' understand the concept. I also wanted to use native libraries
' so that you can dive into the spelling corrector concepts quickly
' without first having to learn other technologies like LINQ etc.
'
' Please feel free to post any comments, suggestions for improvement and any bugs
' you find.

Imports System.Collections.Generic

Public Class SimpleSpellingCorrector

Private Const COMMAASCII As Integer = 44
Private Const ALPHABET As String = "abcdefghijklmnopqrstuvwxyz"
Private nWords As New Dictionary(Of String, Integer)


Public Function getCorrection(ByVal word As String) As String
Dim c As String = String.Empty
Dim maxc As Integer = -1
Dim wc As Integer = 0
Dim candidates As System.Collections.Generic.List(Of String)

'Train the model with word occurences in our "dictionary"
nWords = getModel()

'Choose the most probable word with the shortest edit distance
For ed As Integer = 0 To 2
'If we have found a correction, exit loop
If String.Empty <> c Then
Exit For
End If

'Otherwise, start over
c = String.Empty
wc = 0
maxc = -1
candidates = getCandidates(word, ed)
For Each cd As String In candidates
wc = getWordCount(cd)
If wc > maxc Then
maxc = wc
c = cd
End If
Next cd
Next ed

'If no match is found, just send the same word back!
If String.Empty = c Then
c = word
End If

Return c
End Function

'Get the count of how often the word is found in our "dictionary"
'Return 1 for a "new" word
Private Function getWordCount(ByVal word As String) As Integer
If nWords.ContainsKey(word) Then
Return nWords.Item(word)
Else
Return 1
End If
End Function


'Get the big.txt file from http://norvig.com/big.txt
Private Function getModel() As Dictionary(Of String, Integer)
Dim model As New Dictionary(Of String, Integer)
For Each f As System.Text.RegularExpressions.Match In System.Text.RegularExpressions.Regex.Matches(System.IO.File.ReadAllText("big.txt").ToLower(), "[a-z]+", System.Text.RegularExpressions.RegexOptions.Compiled)
If model.ContainsKey(f.Value) Then
model.Item(f.Value) += 1
Else
model.Add(f.Value, 1)
End If
Next f

Return model
End Function

'Get Candidate words that are at the given edit distance
Private Function getCandidates(ByVal word As String, ByVal edits As Integer) As List(Of String)
Dim c As New List(Of String)
Select Case edits
Case 0
Dim wl As New List(Of String)
wl.Add(word)
c.AddRange(getKnown(wl))
Case 1
c.AddRange(getKnown(getEdits1(word)))
Case 2
c.AddRange(getKnownEdits2(word))
Case Else
c.Add(word)
End Select
Return c
End Function

'Get words that are at an edit distance of 1
Private Function getEdits1(ByVal word As String) As List(Of String)

Dim e1 As New List(Of String)
Dim splits As New List(Of String)

'Create a list of comma separated tuples of all possible ways to split the word
For i As Integer = 0 To word.Length
splits.Add(word.Substring(0, i) & "," & word.Substring(i))
Next i

For Each s As String In splits
'Deletes
If String.Empty <> s.Split(Chr(COMMAASCII))(1) Then
e1.Add(s.Split(Chr(COMMAASCII))(0) & s.Split(Chr(COMMAASCII))(1).Substring(1))
End If

'Transposes
If 1 <> s.Split(Chr(COMMAASCII))(1) Then
e1.Add(s.Split(Chr(COMMAASCII))(0) & c & s.Split(Chr(COMMAASCII))(1).Substring(1))
End If


'Replaces
For Each c As Char In ALPHABET.ToCharArray
If String.Empty <> s.Split(Chr(COMMAASCII))(1) Then
e1.Add(s.Split(Chr(COMMAASCII))(0) & c & s.Split(Chr(COMMAASCII)(1).Substring(1))

End If
Next

'Inserts
For Each c As Char In ALPHABET.ToCharArray
e1.Add(s.Split(Chr(COMMAASCII))(0) & c & s.Split(Chr(COMMAASCII))(1))
Next
Next s

Return e1
End Function

'Get Known words that have edit distance of 2
Private Function getKnownEdits2(ByVal word As String) As List(Of String)
Dim ke2 As New List(Of String)
For Each e1 As String In getEdits1(word)
For Each e2 As String In getEdits1(e1)
If nWords.ContainsKey(e2) Then
ke2.Add(e2)
End If
Next e2
Next e1
Return ke2
End Function

'Get Known words; get rid of unknown words
Private Function getKnown(ByRef words As List(Of String)) As List(Of String)
Dim k As New List(Of String)
For Each w As String In words
If nWords.ContainsKey(w) Then
k.Add(w)
End If
Next w
Return k
End Function

End Class