Brien, P: Fill & Download for Free

Arnab Gowswami strictly does not support any of the political parties although his forefathers were affiliated to almost every political party.His grandfather was a congress leader,father was a BJP leader (earlier belonged to communist party).He has been through the atmosphere of different ideologies.He in his early childhood has seen the worlthless policies of communist party because he has spent much of his early time in Bengal.He holds his anger against the appeasement politics and it can clearly be seen in his interviews.A recent seminar of Arnab with the students of a famous law college, he indicated that Times Now wanted him to manipulate the news and say against the current eastablishments as other news channels do but unfortunately this damaged their own reputation and forced Arnab to resign. Even the last farewell episode of his famous primetime was not allowed to be aired.He is a great fan of Mr.Modi and his policies (but not BJP). Only Arnab as a journalist has got the maximum chances to interview Mr. Modi and question his policies face to face.He was satisfied with his answers (if you want to see his dissatisfaction,watch his interview with RaGa).It comes sometimes to my mind what it would be like if Arnab gets a chance to represent BJP in Rajya sabha.I want him to debate with Derek’o Brien (:-P).

It’s the most liberal metro in India. You will not be denied flats in certain societies because you’re a minority or if you are a non vegetarian. This is something which is very common in Mumbai - the supposed ‘advanced’ city.This is how we celebrate Durga Puja.Here is Christmas being celebrated with the same fervour in Park Street.One of the highest Anglo Indian populations, one of the highest Muslim populations and one of the highest population of people from North East. They coexist with the majority Hindu population.North Easterners prefer Kolkata more than Delhi (developed city huh ?)Kolkata safer than Delhi, say students from northeastA quick stroll through Maidan and Park Street makes it evident how Anglo Indian culture has been preserved and their cuisine is enjoyed by all of Calcutta(p.s. Bar-B-Q). Neil O Brien, Leslie Claudius and Barry O Brien are from Calcutta :3Groups like Bajrang Dal will not make you marry a girl you happen to roam around with.The Durga Puja, Kali Puja, Saraswati Puja are state holidays where people enjoy a lot. By the way Saraswati Puja is also called the Bengali Valentine’s Day :PAny other city in India would attract loads of outrage over this cartoon. Not here though.Go to Park Street-have beef and then go to Maidan and have Pork too, without fearing any retribution.Although there have been some skirmishes (the tasleema nasreem incident) but it’s been largely peaceful. Bengalis take pride in their inclusive culture. It’s a mix of different communities and religions that have given rise to the composite Bengali culture which is far more important than any religion.

There's no reason to get fancy. Given the three main considerations:How often the statistics change?How often the cost calculations change?Reads need to be near real time (< 2s)There's an RDBMS (SQL) approach that doesn't make any assumptions on the first two, while being fast enough for moderately sized data to satisfy the third. This is likely the best solution for someone prototyping an idea, before features are settled down enough to make some cutting decisions that would enable more stuff to be indexed on write.Aggregates down trees are exactly what RDBMSs do with "group by" aggregates. RDBMSs are no good for unbounded relationships like general graphs, but work well with bounded trees (for example, depth <= 128). I've seen the following type of schema work well for large data sets: a unique ID for each node in the complete tree, a (node ID, terminal node ID) table that describes the path to each node in the tree, and the fact tables around each node. The following query computes the aggregates for all nodes in ~500 milliseconds for a full tree with branching factor 5 and depth 8 (~100k nodes), on the analytical DB MonetDB running on dual core box. Adding new nodes to this schema is O(depth) because the path to each node is compressed to each of its ancestors, whereas aggregates are faster because of the de-normalized paths.CREATE TABLE tree (  node_id bigint not null,  end_node_id bigint not null,  PRIMARY KEY (node_id, path_node_id))  -- Random fact table CREATE TABLE stats1 (  node_id bigint not null,  stat int not null,  primary key (node_id)) Generate a summary for all nodes (~500ms for ~100k nodes)SELECT  tree.node_id as node_id,  AVG(stats1.stat) as s1 -- AVG can be any UDF f(int*) FROM tree  INNER JOIN stats1 ON stats1.node_id = tree.end_node_id GROUP BY tree.node_id Generate a summary for a single node (~1ms for ~100k nodes)SELECT  AVG(stats1.stat) as s1 -- again AVG can be any UDF FROM tree  INNER JOIN stats1 ON stats1.node_id = tree.end_node_id WHERE tree.node_id = ${any_node_id} The above queries are data parallelized per node and a good analytical DB will plan them in parallel. Throwing iron at problems like this is a reasonable scale up strategy. Start with server X, then scale to a 32-core machine with 128GB RAM (can get this for under $10k), then shard the tables by node ID and run each query across all the partitions and bring the results together in the application logic. (that order is not necessarily correct for you. e.g. running on EC2 would favor sharding before larger boxes)Another reasonable strategy to bring down the time of the queries gets too much is to invest some time in building your own high performance data structure that solves exactly your problem. In this case, you don't need general Complex Event Processing (CEP) system. A tree with a fast aggregate implementation would do wonders.Here's Scala (programming language) code to run the above queries end to end (though you'll need to create your DB): def main(in: Array[String]) {  Class.forName("nl.cwi.monetdb.jdbc.MonetDriver");  val con = DriverManager.getConnection("jdbc:monetdb://localhost/${your_db_name}", "monetdb", "monetdb");  val s = con.createStatement()   // Clear the tables  s.executeUpdate("DELETE FROM tree")  s.executeUpdate("DELETE FROM stats1")   val treePath = "/home/brien/treetable"  val statsPath = "/home/brien/statstable"  val (n, m) = writeTables(treePath, statsPath)  s.executeUpdate("COPY " + n + " RECORDS INTO tree FROM '" + treePath + "'")  s.executeUpdate("COPY " + m + " RECORDS INTO stats1 FROM '" + statsPath + "'")   for (i <- 0 until 10) {  // Summarize all nodes with a simple average  val time0 = System.nanoTime  val rs = s.executeQuery("""  SELECT tree.node_id as node_id, AVG(stats1.stat) as s1  FROM tree  INNER JOIN stats1 ON stats1.node_id = tree.end_node_id  GROUP BY tree.node_id  """)  while (rs.next()) {  val id = rs.getLong("node_id")  val s1 = rs.getDouble("s1")  if (10 == id) {  printf(">> %d\t%f\n", id, s1)  }  }  rs.close()  val time1 = System.nanoTime   printf("Summarized all nodes in %dms\n", (time1 - time0) / (1000 * 1000))  }   for (i <- 0 until 10) {  // Summarize one node with a simple average  val time0 = System.nanoTime  val rs = s.executeQuery("""  SELECT AVG(stats1.stat) as s1  FROM tree  INNER JOIN stats1 ON stats1.node_id = tree.end_node_id  WHERE tree.node_id = 10  """)  while (rs.next()) {  val s1 = rs.getDouble("s1")  printf(">> %d\t%f\n", 10, s1)  }  rs.close()  val time1 = System.nanoTime   printf("Summarized one node in %dms\n", (time1 - time0) / (1000 * 1000))  }  }   def writeTables(treePath: String, statsPath: String): (Int, Int) = {  val pw = new PrintWriter(new BufferedWriter(new FileWriter(new File(treePath))))  val pws = new PrintWriter(new BufferedWriter(new FileWriter(new File(statsPath))))  val r = new scala.util.Random()   // Insert a full tree of depth D branching B  val D = 8  val B = 5  var id = 0L  var pn = 0  val p = new Array[Long](D)  def btree(d: Int) {  p(d) = id  for (i <- d until D) {  pw.print("%d|%d\n" format (p(i), id))  pn += 1  }  pws.print("%d|%d\n" format (id, r.nextInt(1024 * 1024 * 32)))  id += 1  if (0 < d)  for (i <- 0 until B)  btree(d - 1)  }  btree(D - 1)  pw.close()  pws.close()  printf("Inserted %d nodes\n", id)  (pn, id.toInt)  }