I understand that tokenization criteria are critical to the BLEU scores that one gets, but what I don’t understand is why I am getting inconsistent differences when I compare baseline BLEU scores or Custom BLEU scores. Let me explain.
For instance, I recently trained an English-Danish Custom model with a training set of around 300k segments and a test set or around 2k segments. Once finished, I got a baseline bleu of 34,63 and a Custom bleu of 48,3. Just to double-check these scores, I recalculated the bleu scores of the Model “System Test Results” downloaded from the Custom Translator portal with the Moses tokenizer.perl and the mult-bleu.perl and with the baseline model I get a bleu score of 34,16, but with the Custom model I get 41,57.
How do you explain that with the baseline model I get a very similar score to the one I get from Microsoft, but with the Custom model I get a score about seven points lower than the one I get from Microsoft. The Danish case is just an example, but this is the behavior I observed with several other languages.
Could you please contact us on the custommt support alias? I'd like to see your wordbroken test data and determine why there is such a drastic difference.
Related
I want to predict the occurrences of the word "repeat" in a speech as well as the word's approximate duration. For this task, I'm planning to build a Deep Learning model. I've around 50 positive as well as 50 negative utterances (I couldn't collect more).
Initially I've searched for any pretrained models for keyword spotting, but I couldn't get a good one.
Then I tried Speech Recognition models (Deep Speech), but it couldn't predict the exact repeat words as my data follows Indian accent. Also, I've thought that going for ASR models for this task would be a over-killing one.
Now, I've split the entire audio into chunk of 1 secs with 50% overlapping and tried a binary audio classification in each chunk that is whether the chunk has the word "repeat" or not. For building the classification model, I calculated the MFCC features and build a sequence model on the top of it. Nothing seems to work for me.
If anyone already worked with this kind of task, please provide me with a correct method/resources to build a DL model for this task. Thanks in advance!
So, I'm doing a 4 label x-ray images classification on around 12600 images:
Class1:4000
Class2:3616
Class3:1345
Class4:4000
I'm using VGG-16 architecture pertained on the imageNet dataset with cross-entrpy and SGD and a batch size of 32 and a learning rate of 1e-3 running on pytorch
[[749., 6., 50., 2.],
[ 5., 707., 9., 1.],
[ 56., 8., 752., 0.],
[ 4., 1., 0., 243.]]
I know since both train loss/acc are relatively 0/1 the model is overfitting, though I'm surprised that the val acc is still around 0.9!
How to properly interpret that and what causing it and how to prevent it?
I know it's something like because the accuracy is the argmax of softmax like the actual predictions are getting lower and lower but the argmax always stays the same, but I'm really confused about it! I even let it train for +64 epochs same results flat acc while loss increases gradually!
PS. I have seen other questions with answers and didn't really get an explanation
I think your question already says about what is going on. Your model is overfitting as you have also figured out. Now, as you are training more your model slowly becoming more specialized to the train set and loosing the the capability to generalize gradually. So the softmax probabilities are getting more and more flat. But still it is showing more or less the same accuracy for validation set as still now the correct class has at least slightly more probability than the others. So in my opinion there can be some possible reasons for this:
Your train set and validation set may not be from the same distribution.
Your validation set doesn't cover all cases need to be evaluated, it probably contains similar types of images but they do not differ too much. So, when the model can identify one, it can identify many of them from the validation set. If you add more heterogeneous images in validation set, you will no longer see such a large accuracy in validation set.
Similarly, we can say your train set has images which are heterogeneous i.e, they have a lot of variations, and the validation set is covering only a few varieties, so as training goes on, those minorities are getting less priority as the model yet to have many things to learn and generalize. This can happen if you augment your train-set and your model finds the validation set is relatively easier initially (until overfitting), but as training goes on the model gets lost itself while learning a lot of augmented varieties available in the train set. In this case don't make the augmentation too much wild. Think, if the augmented images are still realistic or not. Do augmentation on images as long as they remain realistic and each type of these images' variations occupy enough representative examples in the train set. Don't include unnecessary situations in augmentation those will never occur in reality, as these unrealistic examples will just increase burden on the model than doing any help.
I have a confusion about the way the LSTM networks work when forecasting with an horizon that is not finite but I'm rather searching for a prediction in whatever time in future. In physical terms I would call it the evolution of the system.
Suppose I have a time series $y(t)$ (output) I want to forecast, and some external inputs $u_1(t), u_2(t),\cdots u_N(t)$ on which the series $y(t)$ depends.
It's common to use the lagged value of the output $y(t)$ as input for the network, such that I schematically have something like (let's consider for simplicity just lag 1 for the output and no lag for the external input):
[y(t-1), u_1(t), u_2(t),\cdots u_N(t)] \to y(t)
In this way of thinking the network, when one wants to do recursive forecast it is forced to use the predicted value at the previous step as input for the next step. In this way we have an effect of propagation of error that makes the long term forecast badly behaving.
Now, my confusion is, I'm thinking as a RNN as a kind of an (simple version) implementation of a state space model where I have the inputs, my output and one or more state variable responsible for the memory of the system. These variables are hidden and not observed.
So now the question, if there is this kind of variable taking already into account previous states of the system why would I need to use the lagged output value as input of my network/model ?
Getting rid of this does my long term forecast would be better, since I'm not expecting anymore the propagation of the error of the forecasted output. (I guess there will be anyway an error in the internal state propagating)
Thanks !
Please see DeepAR - a LSTM forecaster more than one step into the future.
The main contributions of the paper are twofold: (1) we propose an RNN
architecture for probabilistic forecasting, incorporating a negative
Binomial likelihood for count data as well as special treatment for
the case when the magnitudes of the time series vary widely; (2) we
demonstrate empirically on several real-world data sets that this
model produces accurate probabilistic forecasts across a range of
input characteristics, thus showing that modern deep learning-based
approaches can effective address the probabilistic forecasting
problem, which is in contrast to common belief in the field and the
mixed results
In this paper, they forecast multiple steps into the future, to negate exactly what you state here which is the error propagation.
Skipping several steps allows to get more accurate predictions, further into the future.
One more thing done in this paper is predicting percentiles, and interpolating, rather than predicting the value directly. This adds stability, and an error assessment.
Disclaimer - I read an older version of this paper.
I have been working on a super-resolution task. I have this question about determining loss function, So in the case of the task at hand I felt like going with SSIM as a loss function to train my model. I did get a good set of results. Recently I come across perceptual loss function where we compare how a pretrained model looks at the Ground truth(GT) Images and the Super Resolution(SR) Image(Image generated by the model). My question is, I am thinking of using both ((1-SSIM(SR,GT))+Perceptual loss(SR,GT)) loss for backpropagation, so should I use a trade-off parameter between these two losses? if so how can I set up these trade-off parameters? or should I add these losses with equal weights.
PS: the perceptual loss is calculated by finding SSIMs between the feature maps of GT and SR images from the pre-trained model
I have data with integer target class in the range 1-5 where one is the lowest and five the highest. In this case, should I consider it as regression problem and have one node in the output layer?
My way of handling it is:
1- first I convert the labels to binary class matrix
labels = to_categorical(np.asarray(labels))
2- in the output layer, I have five nodes
main_output = Dense(5, activation='sigmoid', name='main_output')(x)
3- I use 'categorical_crossentropy with mean_squared_error when compiling
model.compile(optimizer='rmsprop',loss='categorical_crossentropy',metrics=['mean_squared_error'],loss_weights=[0.2])
Also, can anyone tells me: what is the difference between using categorical_accuracy and 'mean_squared_error in this case?
Regression and classification are vastly different things. If you reimagine this as a regression task than the difference of predicting 2 when the ground truth is 4 will be rated more than if you predict 3 instead of 4. If you have class like car, animal, person you do not care for the ranking between those classes. Predicting car is just as wrong as animal, iff the image shows a person.
Metrics do not impact your learning at all. It is just something that is computed additionally to the loss to show the performance of the model. Here the accuracy makes sense, because this is mostly the metric that we care about. Mean squared error does not tell you how well your model performs. If you get something like 0.0015 mean squared error it sounds good, but it is hard to visualize just how well this performs. In contrast using accuracy and achieving 95% accuracy for example is meaningful.
One last thing you should use softmax instead of sigmoid as your final output to get a probability distribution in your final layer. Softmax will output percentages for every class that sum up to 1. Then crossentropy calculates the difference of the probability distribution of your network output and the ground truth.