cs287
Psets and work related to CS 287.
####Overview
This repo features a number of different language tasks and approaches. Most models are written with the help of the library Torch.
- HW1: Sentiment Analysis (naive Bayes, logistic regression, SVM)
- HW2: POS Tagging (naive Bayes, logistic regression, MLP)
- HW3: Language Modeling (n-gram, neural probabilistic LM)
- HW4: Word Segmentation (n-gram, neural probabilistic LM, RNN/LSTM)
- HW5: Named Entity Recognition (HMM, MEMM, structured perceptron + Viterbi)
####HW 1 Usage
Naive Bayes
th HW1.lua -datafile SST1.hdf5 -classifier nb
Logistic Regression
th HW1.lua -datafile SST1.hdf5 -classifier lr-cross
Limear SVM
th HW1.lua -datafile SST1.hdf5 -classifier lr-hinge
The following hyperparameters can be specified from the command line:
- datafile: the hdf5 file used for training/validation
- classifier: the specific model to run
- alpha: Laplace smoothing coefficient
- lr: learning rate for SGD
- lambda: l2 regularization coefficient
- n_epochs: number of training epochs
- m: mini-batch size
- kfold: number of k-folds for cross-validation
####Primary contributors