# Tag Info

0

You can take a look at this. https://towardsdatascience.com/simple-stock-price-prediction-with-ml-in-python-learners-guide-to-ml-76896910e2ba This is worth a look too. https://www.analyticsvidhya.com/blog/2018/10/predicting-stock-price-machine-learningnd-deep-learning-techniques-python/ Finally, consider this technique, directly below. This is portfolio ...

1

To generate normal random variable, you can use numpy.random.normal where the scale parameter is the standard deviation. To generate multivariate normal random variable, you can use numpy.random.multivariate_normal. If you are an R user, you might like to consider mvrnorm However, I doubt that is what you really want. After all, it is possible to generate ...

2

They should really be more clear about what they mean, but I expect they're using a Laplacian pyramid. As more evidence, they cite: "Denton et al "Deep generative image models using a laplacian pyramid of adversarial networks." The idea is, store a very low resolution copy of your image, and a series of "difference" images. Each difference image tells you ...

1

1) It appears to be nearly working on you exemple 3. Not reaching exactly 0 or 1 happens because of the sigmoid function, that can only reach 0 and 1 asymptotically. For binary prediction, it is quite usual to put a cut-off on the output, here 0.5 will do : X<0.5 => 0 and X>=0.5 => 1. 2) As far as I remember this is in line with the theory : you don't ...

0

NLTK contains options for retrieving brown, treebank corpora with universal tags, instead of their own tagging schemes. nltk.corpus.treebank.tagged_words(tagset='universal') instead of: nltk.corpus.treebank.tagged_words() Similarly: nltk.corpus.brown.tagged_words(tagset='universal') nltk.corpus.nps_chat.tagged_words(tagset='universal') nltk.corpus....

0

Have you tried creating a dataset with 2500 images of cars and ~2500 images of random stuff and trained it first? if so you should try that and see how it works, if it doesnt work you might have to add more images in the "distractor" class. Provide more information about what you find after you do this experiment and may be we can help further?

1

some more information is required to gauge why this might be happening. what is the size of your dataset. what is your learning rate? From the look at the log it seems like your lr is too low, try increasing it until you find a good starting point. Have tou tried varying other hyper-parameters?

1

You could possibly try varying the learning rate or initialise with different weights. Sometimes the optimiser will get stuck in certain local optimums. Alternatively try starting with pretrained weights and perform transfer learning. I found that even when used on a different domain (image to radio signals), it gives good starting accuracies

0

Yes, it does include some: see here for Imagenet 1000, e.g. 985: 'daisy' 987: 'corn' 988: 'acorn',

0

Yes, there are many kinds of plants in Imagenet. image-net.org has a search feature. For example, http://image-net.org/search?q=tree

0

@CapJS, Instead of going ahead with an RNN, which helps you model the dependencies and relationship between your content, I would suggest you to take a look at 1D convolution networks to achieve the classification of activity. There is a nice post talking about something similar : https://machinelearningmastery.com/cnn-models-for-human-activity-...

1

It depends how deep technichally you want to go. You can apply a slight modification of a Survival methods/ cox models that relate the time that passes, before some event occurs, to one or more covariates that may be associated with that quantity of time. Also if you group de features you can make the problem look like as a classical binary classification ...

0

In most cases, one shouldn't retrain a trained network with only the new data. Rather, train the network from scratch with the new and old data. Adding new data and retraining the model just on that new set of data, will probably make your model fit to only that new data, thus forgetting general features from the other data it was trained on. Also, ...

0

How many classes do you have? Bert can handle a high-quality 12k dataset for binary classification. I recommend duplicating your positive test case 4x and sampling a 5k test cases from your negative class. This will give you a balanced dataset. Then implement BERT in google colab using the original GitHub repository from the google Bert team. You can find a ...

2

First, I wouldn't use the word "noisy" here because if you know which instances are "wrong" then these are not noise, they are negative examples. In my opinion "noisy" is when positive and negative cases are mixed together in a way that makes it difficult (or impossible) to distinguish between them. I think this matters because you're more likely to find ...

1

I think LSTM based RNN network having series of image data with a speed label, can be useful for training such model and you can try if you get some data set. Please do share your result as well.

4

The reason you're seeing BERT and its derivatives as benchmarks is probably because it is newer than the other models mentioned and shows state-of-the-art performance on many NLP tasks. Thus, when researchers publish new models they normally want to compare them to the current leading models out there (i.e BERT). I don't know if there has been a study on the ...

0

You can take a purely count-based approach, commonly called maximum likelihood estimation (MLE). Go through the sets and count the frequency of co-occurrence. Then find the most frequent items. Here is a straight-forward (but not very optimized) solution in Python: from collections import defaultdict from itertools import permutations sets = [{1, 2, ...

0

The model would perform better when given similar data as its trained on.i think you should try mixing the two datasets and check the results. To the client, you can show the three results and explain how it is a great product. In place of mAp, show a chart of precision and recall to explain the accuracy.

2

There is plenty of methods to calculate feature importance. I recommend trying two of them LIME and SHAP. I don't want to copy-paste material and tutorial provided by the author so please refer to these two repositories.

1

There is a particular library called as ReduceLROnPlateau, that will reduce the learning rate, based on the factor value you mention. And this seems working good for all problem cases.

2

Let $n$ be a convolutional layer with dimensions $w' \times h' \times c'$. Then each of its $c'$ filters is connected to all $c$ filters (or channels*) of the previous layer. I find it helpful to look at the number of weights here: A single filter of that convolutional layer $n$ with kernel size $k'\times k'$ will have $c \times k' \times k'$ weights. And ...

0

2

You have a big dataset and you get new instances//data every 2 months. First you should select with which data you want to train. Since your data is big and there is the probability than data from 2 years ago is not as relevant as the data from the last month you can consider doing a Roll out// slidding window validation. This way you will only select the ...

0

You need normal data to train on. If you have abnormal instances also, those should be excluded from the training set. Having access to labeled abnormal/normal data is very useful for the validation and testset. Anything that differs from the normal data (as learned by the autoencoder) is considered an anomaly. If you have a lot of labeled abnormal and ...

2

From what I understand - your problem is "sample selection bias" problem. Any kind of pattern to select a subset out of large data may lead to bias. This raises two question. How to choose? Random/stratified random (If you have multiple classes) under sampling to obtain a smaller subset. How big to choose? we can set percentage of undersampling. Reducing ...

2

CNNs like U-Net extract lower level features like edges on lower layers (i.e. the first convolutional layers) and higher level features on higher layers (i.e. convolutional layers closer to the final linear layers). This principle is losely inspired by how visual perception is implemented in the Visual Cortex among humans (and other animals). In a CNN the ...

2

Your model probably overfits. This articles provides an easy to read intro to the topic. As a very first step I suggest to plot your learning curves and look for epochs with lower validation loss. Also, this helps to properly diagnose your model in terms of model capacity. Training a model for less epochs is one way to reduce model capacity and avoid ...

0

It's a quite complex problem and there might be better options for the design (I'm thinking maybe something more specific to times series)... However before that there's a more obvious problem to solve: it seems that you are calculating accuracy on the "Glucose_t" numeric value, right? If yes this is incorrect and that would explain your terrible results: ...

3

Receptive field refers to the number of input pixels that a convolutional filter will operate on. There's a nice distill article about how to calculate receptive field size for your filters (with a nice visualization of receptive field size) and an interactive calculator here if you're only curious about how receptive field size grows with changes to depth ...

1

Your reasoning is perfectly correct. Augmentation is just a process, which helps you cover your domain better. You should only pick operators that help you. Abusing augmentation can definitely mess up your model. It's always good idea to print data at those limits, to check yourself. Try also to think, how data will be acquired on production. Albumentations ...

0

I want to build a deep learning model to predict the next job title when a current title is given. Are there any ways that I could achieve this using some deep learning model? I think that approaching this problem in the classification way (input:- current job embedding, output:- getting next job title as a class) can somewhat be time-consuming (not ...

0

Transfer Learning: for example you want to predict price of article normally we use previous data based on that we design model .while new data came still we use that model for prediction here we are transferring the same model for new task or in general When you learn how to drive a car, you learn a generic skill and you will use these similar set of ...

0

It could be that each capsule is of 2D form (thus 6x6). So each component capsules outputs an 8D vector. The 8D vector is derived by analyzing 8 feature maps propagated from the convolutional layer. So its like placing the feature maps in groups of 8. This means each component capsule represents one capsule therefore in the "dynamic paper", there are ...

0

I don't have enough reputation to comment though, but how many features does your data have? How about using Lasso Regression to select features (using 50 anomaly data and another random 50/100 normal data, assuming that normal data come from the same distribution) and then see when plotting normal vs. abnormal data points, are abnormal points separated ...

0

Andrew Ng's explanation actually covers the YOLOv2 which uses anchor boxes. YOLOv1, which is the paper you linked, does not use anchor boxes so its not exactly the same. They key to understanding how the bounding boxes are formed is to first understand how the output is encoded. To which, I'll recommend this link: https://hackernoon.com/understanding-yolo-...

1

Some things in the field you describe are patentable and some are not. Unfortunately, above novelty and non-obviousness, the current huge hurdle is abstractness. The law on this in the U.S. has changed in the last few years in the direction of making it much easier to shoot down something as abstract. The broader range of processes you intend to cover the ...

1

We use filters mostly to get different features(characteristics) about the object(e.g. image). And pooling we're using to reduce the size and at the same time to save the most significant information of each feature map.

1

Square loss (MSE or SSE) does this. Let $y_i$ be an actual value and $\hat{y}_i$ be its estimated value (prediction). $$SSE = \sum (y_i -\hat{y}_i)^2$$ $$MSE=\dfrac{SSE}{n}$$ Except for numerical issues of doing math on a computer, these are optimized at the same parameter values of your neural network. The squaring is critical. If a prediction is off by ...

0

Notice that from any complete set, you can form an question-answer pair for your task: you can construct an incomplete set (the "question"), for which you know the desired complete set (the "answer"). In particular, given a complete set $S$, you can randomly choose a subset $T$ of $S$ of a particular size, and call $T$ the incomplete set. In this way, from ...

0

The technical term is "recommender system." The biggest problem in recommender systems is the sparsity and size of the data. All algorithms beyond nearest neighbor search for latent features in the same way as PCA. Most common ones are k nearest neighbor and Singular value decomposition. I have seen lstm used to interesting effect though it would follow user'...

0

Adam's answer does seem to make the most sense, however, I am not sure about the second statement "Polluting sequential data with non-sequential information". So recently I trained a character-level LSTM model, in which I just appended a non-sequential feature in the end of the sequential features. The model learned how to differentiate that pretty well. ...

0

you can use some thing like DeepWalk to embed each node so it is straightforward to do any classification on that embedding as it is just a vector.

2

Maybe it's a bit overkill and biased toward my own field (neural machine translation), but you could go with a neural network architecture with self-attention in a masked language model-ish (i.e. BERT) configuration The input to the network would be a fixed-size (40) sequence of discrete symbols meaning whether the element at that position is either present ...

3

You could train an embedding model. Each element would be projected onto a location a vector space based on its co-occurrence with other elements. Then finding similar elements could be done with a nearest neighbor search.

0

Solution: Finally I found the solution myself. I just used another docker image with an older version of tensorflow (2.0.0), and the error disappeared.

0

A major difference between the first and the second model you trained is the size of the data assuming that the model is not pretrained. Increased data, of course, needs increased epochs. According, the batch size must also increase. Batch Size: While training on the smaller dataset, a batch size of 10 yielded better results. The errors were averaged over ...

1

I will try and be as concise as possible. First, let's redefine the way you think about your data points. There can ever only be two types of visits in terms of time. Periodic and Non-Periodic. Let's call each visit an event. Some events could be related to chronic conditions where periodic visits are quite common. Some events could be related to flu, head ...

2

According to your description you can only use similarities between descriptions, and since there's no labelled data it has to be unsupervised. Option 1: heuristic (i.e. ad-hoc unsupervised method). Based on your knowledge of the specifics of the data, implement a function which returns a score representing how similar two descriptions are. For example a ...

0

The question may be too old but I think the BigBadMe answer is not true. As the keras docs said: units: Positive integer, dimensionality of the output space. The number of units actually is the dimension of the hidden state (or the output). For example, in the image below, the hidden state (the red circles) has length 2. The number of units is the ...

Top 50 recent answers are included