On Wed, Mar 17, 2010 at 6:14 AM, Jeff Eastman <j...@windwardsolutions.com>wrote:
> Pallavi Palleti wrote: > >> Hi, >> >> Could some one kindly let me know the significance of instance variable >> "name" in AbstractVector? It is causing problems, when I write a vector to >> file and read and compare with the same vector if the value of "name" is >> null. Because, while writing to file, "name" is set to empty string if it is >> null. So, when we read the vector from the file, it will have different >> value (not null) and asFormatString will have two different values for these >> vectors and so concludes that they are different. >> >> Thanks >> Pallavi >> >> The "name" instance variable was added in MAHOUT-65 along with the > "labelBindings" feature so that e.g. a term vector can retain its term in > its state. I guess the problem you are seeing is an interaction between the > vector Writable implementation - which incorrectly handles null - and the > Json produced by asFormatString. <rant> I've said this before and, not to > belabor the point, using a Json encoding to compare vectors for equality has > a host of related problems, most recently with lazy lengthSquared. If Vector > implemented Printable instead, then asFormatString(bindings) could probably > be crafted to eliminate these problems and be usable for such comparisons. > </rant> I think Pallavi is running into a worse version of this, Jeff - since Writable "rehydrates" a null String as "" as you say, but equals() and even equivalent() in Vector actually do do string compares on the name, VectorWritables do not properly compare with equals before and after serialization with *themselves*, which is a nasty *bug* and needs to get fixed, regardless of Printable / asFormatString work. -jake
