I am trying to reimplement this paper Semantically Conditioned LSTM-based Natural Language Generation for Spoken Dialogue Systems, in which they add a gate to the LSTM cell and change how the state is computed.
How can I do this in tensorflow? Do I need to add a new OP ?
The tf.nn.rnn() and tf.nn.dynamic_rnn() functions accept an argument cell of type tf.nn.rnn_cell.RNNCell. For example you can take a look at the implementation of tf.nn.rnn_cell.BasicLSTMCell (in particular the BasicLSTMCell.__call__() method), which might be a good starting point for your customized LSTM.
Related
I have a problem which involves optimization of actions over time:
Lets assume I have a set of input variables X where each X_i_t has a
value at each point in time t = 0 ... T.
For each point in time, I would like to choose an action a_t of a set of
actions A,
such that a utility function U(a0, ..., a_T) is maximized.
Note, the utility function does not have a closed-form solution and its value depends on the entire sequence of actions a_0 ... a_T.
How would I implement something like this? I am perfectly happy with a keyword I can use to look up relevant literature. I do not need a full solution. - Though if somebody can point me to a python sklearn function which does this, I would definitly not say no...
My first intuition was "logistic regression" but there is no way to assign "correct labels" to an action a_t at time t, since the utility depends on the actions taken earlier and later in the time series.
If you plan to use neural networks with TensorFlow or Pytorch, it will be easy. As long as you can express the function U within the framework and the utility function is reasonably close to being continuous, you can back-propagate the utility to the network. You just ask the optimizer to maximize the utility and that's it.
If the utility function is discrete, it gets tricky, but there are several tricks you might try. One of them is the REINFORCE algorithm (Monte-Carlo policy gradient). Another trick that is getting quite popular is Gubmle softmax that allows sampling of discrete action and propagating the error to the network.
If you plan to use different classifiers (like decision forests or whatever), you might try something based on imitation learning like the SEARN algorithim.
I am coming from Google BERT context (Bidirectional Encoder representations from Transformers). I have gone through architecture and codes. People say this is bidirectional by nature. To make it unidirectional attention some mask is to be applied.
Basically a transformer takes key, values and queries as input; uses encoder decoder architecture; and applies attention to these keys, queries and values. What I understood is we need to pass tokens explicitly rather than transformer understanding this by nature.
Can someone please explain what makes transformer is bidirectional by nature
Bidirectional is actually a carry-over term from RNN/LSTM. Transformer is much more than that.
Transformer and BERT can directly access all positions in the sequence, equivalent to having full random access memory of the sequence during encoding/decoding.
Classic RNN has only access to the hidden state and last token, e.g. encoding of word3 = f(hidden_state, word2), so it has to compress all previous words into a hidden state vector, theoretically possible but a severe limitation in practice. Bidirectional RNN/LSTM is slightly better. Memory networks is another way to work around this. Attention is yet another way to improve LSTM seq2seq models. The insight for Transformer is that you want full memory access and don't need the RNN at all!
Another piece of history: an important ingredient that let us deal with sequence structure without using RNN is positional encoding, which comes from CNN seq2seq model. It would not have been possible without this. Turns out, you don't need the CNN either, as CNN doesn't have full random access, but each convolution filter can only look at a number of neighboring words at a time.
Hence, Transformer is more like FFN, where encoding of word1 = f1(word1, word2, word3), and encoding of word3 = f2(word1, word2, word3). All positions available all the time.
You might also appreciate the beauty which is that the authors made it possible to compute attention for all positions in parallel, through the use of Q, K, V matrices. It's quite magical!
But understand this, you'll also appreciate the limitations of Transformer, that it requires O(N^2 * d) computation where N is the sequence length, precisely because we're doing N*N attention of all words with all other words. RNN, on the other hand, is linear in the sequence length and requires O(N * d^2) computation. d is the dimension of model hidden state.
Transformer just won't write a novel anytime soon!
On the following picture you will see in a really clear way why BERT is Bidirectional.
This is crucial since this forces the model to use information from the entire sentence simultaneously – regardless of the position – to make a good predictions.
BERT has been a clear break through allowed by the use of the notorious "attention is all you need" paper and architecture.
This Bidirectional idea (masked) is different from classic LSTM cells which till now used the forward or the backward method or both but not at the same time.
Edit:
this is done by the transformer. The attention is all you need paper is presenting an encoder-decoder system implementing a sequence to sequence framework. BERT is using this Transformer (sequence to sequence Bidirectional network) to do other NLP task. And this has been done by using a masked approach.
The important thing is: BERT uses Attention but Attention has been done for a translation and as such do not care for Bidirectional. But remove a word and you have Bidirectional.
So why BERT now?
well the Transformer is the first transduction model relying
entirely on self-attention to compute representations of its input and output without using sequencealigned RNNs or convolution. Meaning that this model allows a sentence Embedding far more effective than before. In fact, RNN based architectures are hard to parallelize and can have difficulty learning long-range dependencies within the input and output sequences. SO break through in architecture AND the use of this idea to train a network by masking a word (or more) leads to BERT.
Edit of Edit:
forget about the scale product, it's the inside the Attention which is inside A multi head attention itself inside the Transformer: you are looking to deep. The transformer is using the entire sequence every time to find the other sequence (In case of BERT it's the missing 0.15 percentage of the sentence) and that's it. The use of BERT as a language model is realy a transfer learning (see this)
As stated in your post, unidirectional can be done with a certain type of mask, bidirec is better. And it is used because the go from a full sentence to a full sentence but not the way classic Seq2seq is made (with LSTM and RNN) and as such can be used for LM.
BERT is a bidirectional transformer whereas the original transformer (Vaswani et al., 2017) is unidirectional. This can be shown by comparing the masks in the code.
Original Transformer
Tensorflow's tutorial is a good reference. The look_ahead_mask is what makes the Transformer unidirectional.
def create_look_ahead_mask(size):
mask = 1 - tf.linalg.band_part(tf.ones((size, size)), -1, 0)
return mask # (seq_len, seq_len)
If you trace the code, you can find the look_ahead_mask is applied to the attention_weights in the decoder. Basically each row in the attention_weights represents a attention query for token at certain position (the first row -> first token position; the second row -> second token position etc.). And the look_ahead_mask blacks out the tokens appear after this position in the decoder so it does not see the "future". In that sense, the decoder is unidirectional analogous to unidirectional in an RNN.
BERT
On the other hand, if you check the original BERT implementation (also in Tensorflow). There's only an optional input_mask applied to the entire BertModel. And if you follow the README on pre-training the model and run create_pretraining_data.py, you will observe that the input_mask is only used for padding the input sequence so for short sentences the unused tokens are ignored. Thus, attention in BERT can be applied to both the "past" and the "future" of a given token position. In that sense, the encoder in BERT is bidirectional analogous to bidirectional in an RNN.
I know this is an old post but for anyone coming back to this:
Adding to what Hai-Anh Trinh said, Transformers aren't 'bi-directional', it would be better to call them "omni-directional". Because of their self-attention method, they are able to consider every single word at the same time, simultaneously.
BERT on the other hand is "deeply bidirectional". This is because of the masked language model(MLM) pre-training objective that is used in BERT. (there are a lot of resources online, I can link some if need be)
It's easy to get confused so don't worry about it.
(https://arxiv.org/pdf/1810.04805.pdf; link to the original BERT paper)
(https://arxiv.org/pdf/1706.03762.pdf; link to the original Transformer paper)
I have around 300 features and I want to find the best subset of features by using feature selection techniques in weka. Can someone please tell me what method to use to remove redundant features in weka :)
There are mainly two types of feature selection techniques that you can use using Weka:
Feature selection with wrapper method:
"Wrapper methods consider the selection of a set of features as a search problem, where different combinations are prepared, evaluated and compared to other combinations. A predictive model us used to evaluate a combination of features and assign a score based on model accuracy.
The search process may be methodical such as a best-first search, it may stochastic such as a random hill-climbing algorithm, or it may use heuristics, like forward and backward passes to add and remove features.
An example if a wrapper method is the recursive feature elimination algorithm." [From http://machinelearningmastery.com/an-introduction-to-feature-selection/]
Feature selection with filter method:
"Filter feature selection methods apply a statistical measure to assign a scoring to each feature. The features are ranked by the score and either selected to be kept or removed from the dataset. The methods are often univariate and consider the feature independently, or with regard to the dependent variable.
Example of some filter methods include the Chi squared test, information gain and correlation coefficient scores." [From http://machinelearningmastery.com/an-introduction-to-feature-selection/]
If you are using Weka GUI, then you can take a look at two of my video casts here and here.
I am learning about Hidden Markov Models for classifying motion in a sequence of t image frames.
Assume I have m dimensions of feature from each frame. Then I cluster it into a symbol (for observable symbol). And I create k different HMM model for k class.
Then, how do I determine the number of hidden states for each model to optimise prediction ?
Btw, is my approach correct? If I misunderstood how to use it, please correct me:)
Thanks :)
"is my approach already correct?"
Your current approach is correct. I have done the same thing some weeks ago and had asked the same questions. I have built a gesture recognition tool.
You say you have k classes you want to recognize, so yes, you will train k HMM. For each HMM you run Forward algorithm and receive P(HMM|observation) for each hidden markov model (alternatively Viterbi decoding is also possible). Then you take the one with the highest probability.
It's also correct to see the m-dimensional feature vector as a single observation symbol. Depending on what your vector looks like, you might want to use a continuous hidden markov model or a discrete hidden markov model. Working with discrete ones is often easier and easier to train with little training data. So in case your feature vector space is continuous, you might want to consider discretization to make all values discrete (e.g. through uniform classes).
The question about discreteness is: How many classes of observations will you have?
"how to determine the number of hidden state for each model to get optimal prediction?"
However, I cannot fully answer your actual question about the number of hidden states. From what I have been taught in other areas, it seems like it's a lot of benchmarking and testing. E.g. in speech recongition we use 3 HMM states for each phonem (human sound), because sounds sound different at the beginning, in the middle and at the end. And then each different phonem gets one triple. But that was of course engineering.
In my own application I have thought like this: I wanted to define gestures and associate them with directions. Like open_firefox = [UP, RIGHT]. So I decided to use four hidden states for all four directions.
I guess finding out the best number of states is a lot about engineering and trying out different things.
I would like to use a supervised machine learning algorithm to predict a binary function (true or false) for a set of sentences based on the presence or absence of words in the sentences.
Ideally, I would like to avoid having to hardcode the set of words used to decide on the output so that the algorithm automatically learns which words are (together ?) most likely to trigger specific outputs.
http://shop.oreilly.com/product/9780596529321.do (Programming Collective Intelligence) has a nice section in chapter 4 titled "Learning From Clicks" which describes how to do this by using 1 layer of hiden nodes in a neural network with one new hidden node for each new combination of input words.
Similarly, it is possible to create a feature for each word in the training data set and train pretty much any classic machine learning algorithm using these features. Adding new training data will generate new features which will require me to re-train the algorithm from scratch.
Which brings me to my questions:
is it actually a problem if I have to retrain everything from scratch whenever the training data set is extended ?
what kind of algorithm would more experience machine learning users recommend to use for this kind of problem ?
what criteria should I use in picking an algorithm versus another ? (other than actually trying them all and see which perform better with precision/recall metrics)
if you have worked on similar problems, what about extending the features with 2-grams (1 if a specific 2-gram is present, 0 if not) ? 3-grams ?
You could look into the general area of topic modelling if you want to find words which are generally found together.
The most simple approach would be to use latent semantic analysis ( http://en.wikipedia.org/wiki/Latent_semantic_analysis ), which is just applying SVD to a term document matrix. You'd then need to do some additional post hoc analysis to fit this to your particular outcome.
A more involved, and much more complex approach would be to use latent dirichlet allocation ( http://en.wikipedia.org/wiki/Latent_Dirichlet_allocation )
In terms of just adding new features (words) that is fine as long as you are going to retrain. You can also use TF/IDF to give that particular word a value when representing the matrix (Instead of just a 1 or 0).
I don't know what programming language you are trying to do this in, but I know there are libraries out there in Java and Pythont hat do all of the above.