I am using the tensorflow website's tutorial for time series modeling on my own data. The models are pretty good, but my target value is always positive and sometimes the model predicts a negative value. Is there a way to clip the model predictions to a range?
Related
Focal Loss given in Tensorflow is used for class imbalance. For Binary class classification, there are a lots of codes available but for Multiclass classification, a very little help is there. I ran the code with One Hot Encoded target variables of 250 classes and it gave me results without any error.
y = pd.get_dummies(df['target']) # One hot encoded target classes
model.compile(
optimizer="adam", loss=tfa.losses.SigmoidFocalCrossEntropy(), metrics= metric
)
I just want to know whoever wrote this code or someone having enough knowledge of this code, can it be used be used for Multiclass Classification. If no then how come it did not give me errors, instead better results than CrossEntropy. Also, in other implementations like this one, the value of alpha has to be given for every class but just one value in Tensorflow's implementations.
What is the correct way to use this?
Some basics first.
Categorical Crossentropy is designed to incentivize a model a model to predict 100% for the correct label. It was designed for models that predict single-label multi-class classification - like CIFAR10 or Imagenet. Usually these models finish in a Dense layer with more than one output.
Binary Crossentropy is designed to incentivize a model to predict 100% if the label is one, or, 0% is the label is zero. Usually these models finish in a Dense layer with exactly one output.
When you apply Binary Crossentropy to a single-label multi-class classification problem, you are doing something that is mathematically valid but defines a slightly different task: you are incentivizing a single-label classification model to not only get the true label correct, but also minimize the false labels.
For example, if your target is dog, and your model predict 60% dog, CCE doesn't care if your model predicts 20% cat and 20% French horn, or, 40% cat and 0% French horn. So this is aligned with a top-1 accuracy concept.
But if you take that same model and apply BCE, and your model predictions 60% dog, BCE DOES care if your models predict 20%/20% cat/frenchhorn, vs 40%/0% cat/frenchhorn. To put it in precise terminology, the former is more "calibrated" and so it has some additional measure of goodness. However, this has little correlation to top-1 accuracy.
When you use BCE, presumably you are wasting the model's energy to focus on calibration at the expense of top-1 acc. But as you might have seen, it doesn't always work out that way. Sometimes BCE gives you superior results. I don't know that there's a clear explanation of that but I'd assume that the additional signals (in the case of Imagenet, you'll literally get 1000 times more signals) somehow creates a smoother loss value that perhaps helps smooth the gradients you receive.
The alpha value of focal loss additionally penalizes very wrong predictions and lessens the penalty if your model predicts something close to the right answer - like predicting 90% cat if the ground truth is cat. This would be a shift from the original definition of CCE, based on the theory of Maximum Likelihood Estimation... which focuses on calibration... vs the normal metric most ML practitioners care about: top-1 accuracy.
Focal loss was originally designed for binary classification so the original formulation only has a single alpha value. The repo you pointed to extends the concept of Focal Loss to single-label classification and therefore there are multiple alpha values: one per class. However, by my read, it loses the additional possible smoothing effect of BCE.
Net net, for the best results, you'll want to benchmark CCE, BCE, Binary Focal Loss (out of TFA and per the original paper), and the single-label multi-class Focal Loss that you found in that repo. In general, those the discovery of those alpha values is done via guess & check, or grid search.
There's a lot of manual guessing and checking in ML unfortunately.
When dealing with time series forecasting, I've seen most people follow these steps when using an LSTM model:
Obtain, clean, and pre-process data
Take out validation dataset for future comparison with model predictions
Initialise and train LSTM model
Use a copy of validation dataset to be pre-processed exactly like the training data
Use trained model to make predictions on the transformed validation data
Evaluate results: predictions vs validation
However, if the model is accurate, how do you make predictions that go beyond the end of the validation period?
The following only accepts data that have been transformed in the same way as the training data, but for predictions that go beyond the validation period, you don't have any input data to feed to the model. So, how do people do this?
# Predictions vs validation
predictions = model.predict(transformed_validation)
# Future predictions
future_predictions = model.predict(?)
To predict the ith value, your LSTM model need last N values.
So if you want to forecast, you should use each prediction to predict the next one.
In other terms you have to loop over something like
prediction = model.predict(X[-N:])
X.append(prediction)
As you can guess, you add your output in your input that's why your predictions can diverge and amplify uncertainty.
Other model are more stable to predict far future.
You have to break your data into training and testing and then fit your mode. Finally, you make a prediction like this.
future_predictions = model.predict(X_test)
Check out the link below for all details.
Time-Series Forecasting: Predicting Stock Prices Using An LSTM Model
I'm developing recommender system using NCF in somewhat modified way.
My circumstance is that data for prediction occasionally includes data used in training.
For example, My training set is 100000rows. And by negative sampling, some unobserved datas are added to training set.
And I want to predict all the unobserved one from the trained model. Then some datas from negative sampling is intersection of train data and predict data.
Will this cause any problem?
Should I remove unobserved data from negative sampling in predict data?
I am trying to create a script that is able to evaluate a model on lfw dataset. As a process, I am reading pair of images (using the LFW annotation list), track and crop the face, align it and pass it through a pre-trained facenet model (.pb using tensorflow) and extract the features. The feature vector size = (1,128) and the input image is (160,160).
To evaluate for the verification task, I am using a Siamese architecture. That is, I am passing a pair of images (same or different person) from two identical models ([2 x facenet] , this is equivalent like passing a batch of images with size 2 from a single network) and calculating the euclidean distance of the embeddings. Finally, I am training a linear SVM classifier to extract 0 when the embedding distance is small and 1 otherwise using pair labels. This way I am trying to learn a threshold to be used while testing.
Using this architecture I am getting a score of 60% maximum. On the other hand, using the same architecture on other models (e.g vgg-face), where the features are 4096 [fc7:0] (not embeddings) I am getting 90%. I definitely cannot replicate the scores that I see online (99.x%), but using the embeddings the score is very low. Is there something wrong with the pipeline in general ?? How can I evaluate the embeddings for verification?
Nevermind, the approach is correct, facenet model that is available online is poorly trained and that is the reason for the poor score. Since this model is trained on another dataset and not the original one that is described in the paper (obviously), verification score will be less than expected. However, if you set a constant threshold to the desired value you can probably increase true positives but by sacrificing f1 score.
You can use a similarity search engine. Either using approximated kNN search libraries such as Faiss or Nmslib, cloud-ready similarity search open-source tools such as Milvus, or production-ready managed service such as Pinecone.io.
I'm not sure if my understanding is correct but...
While training a seq2seq model, one of the purpose I want to initiated a set of pre-trained fasttext weights in the embedding layers is to decrease the unknown words in the test environment (these unknown words are not in training set). Since pre-trained fasttext model has larger vocabulary, during test environment, the unknown word can be represented by fasttext out-of-vocabulary word vectors, which supposed to have similar direction of the semantic similar words in the training set.
However, due to the fact that the initial fasttext weights in the embedding layers will be updated through the training process (updating weights generates better results). I am wondering if the updated embedding weights would distort the relationship of semantic similarity between words and undermine the representation of fasttext out-of-vocabulary word vectors? (and, between those updated embedding weights and word vectors in the initial embedding layers but their corresponding ID didn't appear in the training data)
If the input ID can be distributed represented vectors extracted from pre-trained model and, then, map these pre-trained word vectors (fixed weights while training) via a lookup table to the embedding layers (these weights will be updated while training), would it be a better solution?
Any suggestions will be appreciated!
You are correct about the problem: when using pre-trained vector and fine-tuning them in your final model, the words that are infrequent or hasn't appear in your training set won't get any updates.
Now, usually one can test how much of the issue for your particular case this is. E.g. if you have a validation set, try fine-tuning and not fine-tuning the weights and see what's the difference in model performance on validation set.
If you see a big difference in performance on validation set when you are not fine-tuning, here is a few ways to handle this:
a) Add a linear transformation layer after not-trainable embeddings. Fine-tuning embeddings in many cases does affine transformations to the space, so one can capture this in a separate layer that can be applied at test time.
E.g. A is pre-trained embedding matrix:
embeds = tf.nn.embedding_lookup(A, tokens)
X = tf.get_variable("X", [embed_size, embed_size])
b = tf.get_vairable("b", [embed_size])
embeds = tf.mul(embeds, X) + b
b) Keep pre-trained embeddings in the not-trainable embedding matrix A. Add trainable embedding matrix B, that has a smaller vocab of popular words in your training set and embedding size. Lookup words both in A and B (and if word is out of vocab use ID=0 for example), concat results and use it input to your model. This way you will teach your model to use mostly A and sometimes rely on B for popular words in your training set.
fixed_embeds = tf.nn.embedding_lookup(A, tokens)
B = tf.get_variable("B", [smaller_vocab_size, embed_size])
oov_tokens = tf.where(tf.less(tokens, smaller_vocab_size), tokens, tf.zeros(tf.shape(tokens), dtype=tokens.dtype))
dyn_embeds = tf.nn.embedding_lookup(B, oov_tokens)
embeds = tf.concat([fixed_embeds, dyn_embeds], 1)