Age | Commit message (Collapse) | Author | Files | Lines | |
---|---|---|---|---|---|
2015-04-07 | Report: update test resultsHEADmaster | Peter Wu | 2 | -11/+42 | |
2015-04-07 | Print resulting sentence and score for debugging | Peter Wu | 1 | -2/+7 | |
2015-04-04 | Extend report | Peter Wu | 3 | -15/+101 | |
2015-04-04 | Fix bugs when looking for more than one typo | Peter Wu | 1 | -3/+7 | |
I thought that arraycopy had memcpy semantics, but oh gawd I was wrong. The other bug is that the best result could be overwritten by a worse one since the wrong result was checked. | |||||
2015-04-04 | Split NO_PEACH to DEBUG_PRINTS | Peter Wu | 2 | -4/+5 | |
2015-04-04 | Adjust probability for larger typos count | Peter Wu | 1 | -2/+35 | |
2015-04-04 | Debugging tweaks, be less verbose unless requested | Peter Wu | 3 | -10/+24 | |
2015-04-04 | Labda interpolation works bad | Peter Wu | 1 | -11/+3 | |
It favors large unigram probabilities too much. | |||||
2015-04-04 | Save readonly state across calls | Peter Wu | 1 | -28/+56 | |
Can be cleaned up a bit... but functionality first now. | |||||
2015-04-04 | Changes must always be better than the latest modification | Peter Wu | 1 | -3/+5 | |
2015-04-04 | Do not count unigrams twice | Peter Wu | 1 | -18/+14 | |
2015-04-03 | Report: update with non-greedy approach and more | Peter Wu | 1 | -67/+126 | |
2015-04-03 | Correctly check read-only words | Peter Wu | 1 | -3/+9 | |
2015-04-03 | Non-existing words should not kill the overall score | Peter Wu | 1 | -0/+5 | |
2015-04-03 | Consider changes to ngram probabilties in the context | Peter Wu | 1 | -6/+32 | |
Finally "hme locations" gets corrected to "home locations"! | |||||
2015-04-03 | Save all candidate modifications in one place | Peter Wu | 1 | -42/+48 | |
Also improve debugging to print the difference in score and cache the evaluation result (sentence score). Behavioral difference: now the best score is determined by the whole sentence instead of one word. This means that multiple faulty words will always result in a rating of 0.... | |||||
2015-04-03 | Fix ngram PREFIX condition, improve ngram prob debug | Peter Wu | 2 | -5/+12 | |
2015-04-02 | Further attmept to improve score: drop .5 correction | Peter Wu | 1 | -6/+9 | |
2015-04-02 | Improve ngram score debugging details | Peter Wu | 1 | -6/+25 | |
2015-04-02 | Improve debugging | Peter Wu | 1 | -5/+14 | |
2015-04-02 | Implement generic ngram counting and probability calculation | Peter Wu | 2 | -27/+66 | |
TODO improve smoothing. | |||||
2015-04-02 | Remove broken greedy approach, fix corrections limit | Peter Wu | 1 | -38/+18 | |
2015-04-02 | quick nongreedy | Peter Wu | 1 | -7/+62 | |
2015-04-02 | Merge branch 'master' of git.lekensteyn.nl:tue/2ID90-AI/assignment4 | Wilco | 3 | -24/+19 | |
2015-04-02 | several fixes in report | Wilco | 2 | -13/+13 | |
2015-04-02 | Debug and lambda tweaks | Peter Wu | 1 | -2/+6 | |
Damn, greedy does not work.... | |||||
2015-04-02 | use max instead of sum | Peter Wu | 1 | -4/+3 | |
2015-04-02 | fix output printing | Peter Wu | 1 | -1/+1 | |
2015-04-02 | bigram smooth tweak | Peter Wu | 1 | -1/+2 | |
2015-04-02 | Split lambda p | Peter Wu | 1 | -6/+8 | |
2015-04-02 | Accidentally added two times debug | Peter Wu | 1 | -11/+0 | |
2015-04-02 | Improved score debugging | Peter Wu | 1 | -0/+27 | |
2015-04-02 | Score debugging | Peter Wu | 1 | -0/+13 | |
2015-04-02 | run.sh: allow classpath override, answer only mode | Peter Wu | 2 | -5/+13 | |
2015-04-02 | Consider the probability of the whole sentence | Peter Wu | 1 | -15/+40 | |
2015-04-02 | SpellCorrector: look for differences in likelihood | Peter Wu | 1 | -0/+4 | |
2015-04-02 | Calculate probability for unigram | Peter Wu | 1 | -1/+1 | |
Find L * P(word) instead of L * #word. | |||||
2015-04-02 | Fix probability calculations | Peter Wu | 2 | -36/+45 | |
Replace monkey patching by an actual implementation that correctly combines channel probability (language model), prior and interpolation. | |||||
2015-04-02 | Support continuously processing stdin | Peter Wu | 2 | -1/+20 | |
2015-04-01 | Use Kerninghans confusion matrix | Peter Wu | 1 | -1573/+1261 | |
2015-04-01 | SpellChecker: FUBARED | Peter Wu | 1 | -31/+34 | |
This is the submitted version, it contains random attempts and a fix (actually not that sure that this is the submitted version...). The Double.MIN_VALUE fix is probably missing. | |||||
2015-04-01 | Netbeans: enable assertions on run | Peter Wu | 1 | -1/+1 | |
2015-04-01 | Rate multi n-grams higher | Peter Wu | 1 | -4/+16 | |
2015-04-01 | Set true for peach | Peter Wu | 1 | -1/+1 | |
2015-04-01 | Count unigrams | Peter Wu | 1 | -0/+9 | |
2015-04-01 | Initial check-in of report | Peter Wu | 1 | -0/+200 | |
2015-04-01 | Initial attempt at rating sentences | Peter Wu | 1 | -6/+145 | |
2015-04-01 | Do not include unchanged words | Peter Wu | 2 | -1/+6 | |
2015-04-01 | Remove unused calculateChannelModelProbability | Peter Wu | 1 | -8/+0 | |
Channel probability is already calculated during getCandidateWords. | |||||
2015-04-01 | Calculate noisy channel probability | Peter Wu | 2 | -0/+29 | |