chrispiech / deepknowledgetracing Goto Github PK

View Code? Open in Web Editor NEW

268.0 15.0 89.0 3.55 MB

source code for the paper Deep Knowledge Tracing

Lua 100.00%

deepknowledgetracing's Introduction

DeepKnowledgeTracing

source code for the paper Deep Knowledge Tracing. http://stanford.edu/~cpiech/bio/papers/deepKnowledgeTracing.pdf

At the moment the code doesn't include LSTM model, only RNN (I will upload LSTM when I get a chance).

Questions people have asked me:

Q. How do you handle multiple students during training? It looks like sequences (ie. data of different 
students) of different length are padded to the same length.
A. Correct. This is very important for training speed.

Q. Does the training code have a termination condition?
A. No. I save a copy of the model each epoch and let training run until I feel like terminating it. You can 
start training 
from any saved model (where you left off).

deepknowledgetracing's People

Contributors

Stargazers

Watchers

Forkers

beardeer ml-lab kdjyss benjamesbabala adamist521 hycis skallumadi mrsaibot vruvora codeaudit codeviii xuchongbo wsu4 asmith26 webbp wpmarinho liaopan sandragreiss xuewenfei kevinhuuu spolyak tanghuiyu felixmaximilian varunudacity wangguojie lkhcnn fengchangfight sunbjt sojvai read-papers hsharma83 kumarabhinav04 knowledgehacker lidhcs hs97 snippercahe qqpann cloudspurs renamori donaldcwl songlielie windshield-rider xianglinhe chloeling gavinwb b2220333 cutekane sun80231314 zxcej vivian-nankai 54zy xmzhao danrann fulna converseg xiangrongxu tengben1989 yandexuanxuan chuhai-lin ankitamandal badicst huang-jinyun kumareshsingh yuanhuan177 adldotori lewisbakkero qiaozhqz forsakeljk chunyi0219 moduxiaohunhun annajiat rollbear-bot llliyin shawn-zwj zhengbo-ever superjiang1998 xingyezn liying233 jiang-zn algorithmlover2016 jaemyoung kingsjr26 lyquentxy cdrhim sharetech fjnujerry0 clabra zhouzijie01 hiepkhachbk

deepknowledgetracing's Issues

About dataAssist.lua

Hello, I get little confused in the dataAssist.lua, row 103 "student['questionId'][i] = tonumber(id) + 1". Why plus 1? you store the id of the question in the questionId of the student,why plus one to the id?

how to solve this problem

Can you explain the meaning of the contents in data file named as "naive_c5_q50_s4000_v12_info.txt"？

There are 4 columns in these files, I guess the third column means the concept, but I don't know another three, can you tell me it? Thank you very much

How transform original assistements datasets into experiment dataset in paper?

https://sites.google.com/site/assistmentsdata/home/assistment-2009-2010-data
This is the website of original assistements datasets.But how author transform it into experiment dataset ? Someone can tell me? Thanks~

input data

Hi,

Thanks for sharing your code. But I am a little bit confused about the input data of the model, e.g. the assignments. Could you provide some explanation for it?

Forgive me being a newbie about this, but isn't the knowledge tracing kind of like movie rating prediction in collaborative filtering, where you are asked to predict unseen ratings given some historical ratings of users?

Is there an appoximate bias in AUC calculation

Notice the way AUC is calculated is by approximating to the true one where the code is

for i,p in ipairs(allPredictions) do
		if(p['truth'] == 1) then
			truePositives = truePositives + 1
		else
			falsePositives = falsePositives + 1
		end

		local guess = 0
		if(p['pred'] > 0.5) then guess = 1 end
		if(guess == p['truth']) then correct = correct + 1 end

		local fpr = falsePositives / totalNegatives
		local tpr = truePositives / totalPositives
		if(i % 500 == 0) then
			if lastFpr ~= nil then
				local trapezoid = (tpr + lastTpr) * (fpr - lastFpr) *.5
				auc = auc + trapezoid
			end	
			lastFpr = fpr
			lastTpr = tpr
		end
		if(recall == 1) then break end
	end

Wondering whether this appoximating calculation may affect the auc value.

How is Synthetic created?

Hello @chrispiech ,

Thanks for the code.
Is there the code used to generate syntheteic(simulated-5)?

In your paper, I read that there are 50 exercises and 5 concepts, but does each exercise have all 5 concepts as hidden parameter, or only one concept it requires?
I am trying to reproduce simulated data, but without further information I can't create one that can be trained well.

Thanks in advance,

How to run this Lua code?

Has anyone successfully run this code?
I'm a Lua novice, and I tried to run the code on Ubuntu 20.04.
Torch has been installed on my Linux.

about computing conditional influence between exercises

Hi, Could you give more details in computing conditional influence between exercises? does the influence probability between exercises relate to the predict probability of ouput layer? thanks!