Limit the vocabulary size using word frequencies + remove LR decay#43
Open
vikram-gupta wants to merge 1 commit intomacournoyer:masterfrom
Open
Limit the vocabulary size using word frequencies + remove LR decay#43vikram-gupta wants to merge 1 commit intomacournoyer:masterfrom
vikram-gupta wants to merge 1 commit intomacournoyer:masterfrom
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hi @macournoyer ,
This is a small PR heavily inspired by the code changes done by @chenb67 in her fork and the discussions with you.
The training is going on with these changes on a small dataset. Will update the results soon.
Have tried to keep the code changes similar to @chenb67 , so that we do not face too many conflicts while merging.
-- Update --
Since the dataset and vocabulary was small, some of the responses are not as good as before, but overall the results are satisfactory ! The perplexity kept on decreasing with epoch and came down to around 2.76 after 50 epochs.
Results:
th train.lua --cuda --dataset 25000 --hiddenSize 1000 --maxVocabSize 10000
you> Hi!
neuralconvo> Yes?
you> Hello?
neuralconvo> Hey.
you> How are you?
neuralconvo> I need to know you, hildy.
you> What's your name?
neuralconvo> Smith.
you> How so?
neuralconvo> He was uptown when i it, but saw i don't have to
you> What is the purpose of life?
neuralconvo> My baby daughter.
you> Are you intelligent?
neuralconvo> No.
you> Are you a machine?
neuralconvo> Yeah.
you> Are you human?
neuralconvo> No sir.
you> What are you?
neuralconvo> It's the maintenance entrance.
you> Do you plan on taking over the world?
neuralconvo> Yes.
you> Hi !
neuralconvo> Yes?
you> How is your brother?
neuralconvo> Terrible.
you> Are you married?
neuralconvo> No sir.