New answers tagged

0

The main problem with very little data is that it's almost impossible to know how representative the sample is. Some people would even say that less 20-30 data points cannot be representative of anything. Every single data point can have a huge impact on any model, so any prediction has a huge margin of error. If one is going to train a model from a tiny ...


0

The test data is a 3 step sequence data i.e. 70,80,90. Following will be the flow - 70 will reach the 1st LSTM layer i.e. to the all 100 in parallel It will also come back by the Recurrent weight to be multiplied with the next seq. 80 will reach and add up with the Recurrent of 70. Similarly, 90 will pass through the 1st LSTM layer. All the seq_step is held ...


0

You need two backbones: for time series, use LSTM/GRU and for non-time series, use 1D CNN or Linear layers. Once you get the final embeddings from both of them, concatenate the embeddings and finally feed to the classification layer.


0

The Solution that I have found: y <- ts(data$VALORE, frequency = 24, start = 1) train <- ts(y[1:12264], frequency=24) #70% val <- ts(y[12265:17520], frequency=24) #30%


0

So the question is truly about how to aggregate the data to answer your question, and that has little to do with the fact that you are working with geolocation data, and a lot more with the question you are trying to answer. Let me make some examples: you are a marine scientist who tagged a single whale you are following daily in her journey across our ...


0

Perhaps your repressor is for one time step behind. For e.g. if you are predicting tomorrow's stock price, then the repressor say market volume can be from today. The other thing you can do is build a separate model that estimates the regressor for the next day and then use that in prophet. Be aware that your regressor is then becomes imperfect since it ...


1

I don't know exactly whether it is useful for your case, but you can use last day of the week, so coordinates of the object at end of the week. I copied your example data to a text file, read it with Pandas, and resampled data from daily to weekly by getting coordinates of the last day of each week. df = pd.read_csv('untitled.txt', ...


1

Better to use Collab indeed. Kaggle also provides notebooks with 38h GPU and also 30 hours of TPU per week you might want to have a look at that as well (plus Kaggle allows you to use your GCP credentials so you can link private google cloud storage buckets to your Kaggle notebook). On Kaggle you will also find plenty of public notebooks that can be of great ...


0

It depends how you want to cluster the data, but here are some options.... FEATURE ENGINEERING You could, for example, completely ignore the timestamps and just seek to cluster the different modes of operation are based on the magnitude of the feature alone. Here, simply extract the values into a new feature list. Otherwise, you have to consider what is ...


0

LSTM seeks inputs having the same shape. Hence, you may first consider defining a sequence length and then padding the sequences whose lengths are less than the defined sequence length.


0

The reason you get predictions without an exact time is because that is how models are trained. They are not trained to predict the exact time but a window for it. The reason a model is not trained to predict exact time is because it introduces a lot of problems starting with data imbalance and the huge number of classes it would introduce. Also, DL/AI is ...


0

I'm unclear whether transformers are the best tool for time series forecasting. Transformers at the end of the day are just the latest in a series of sequence-to-sequence models with an encoder and decoder. This means that transformers change something to something else. With time series you aren't changing something to something else, you're trying to find ...


1

You can use RNN architectures like LSTM, and GRU. RNNs take input vectors in each time step, so you can add your extra data to input vector. Your input shape will be batch_size x sequence length x num_of_features.


0

You can use timedelta function from datetime to achieve this, starting from a known date. Example code can be something like, from datetime import datetime, timedelta start = datetime.strptime('2021-01-01', '%Y-%m-%d') all_dates = [start + timedelta(x) for x in range(10)] print(all_dates) Replace the range(10) above with the sequence of the actual days ...


1

Would something like this work? I simply add an extra column that indicates the row number (which is later used as the x-axis) to make sure all values are displayed as a new point instead of plotting on top of each other for the same day. I then specifiy the custom x ticks and labels by selecting the first row for each day and get the row number (which ...


0

Looking at your data - the easiest way is to create a Last-N Days hourly average of the binary indicator - and then use a threshold (based on experimentation) to binarize it. e.g. if your Last 10 Day hourly average looks like this: 0, 0, 0.6, 0.8, 0.9, 1, 0.9, 0.7, 0, 1, 1, 1, 0 Then, a threshold of 0.8 to binarize would result in the following: 0, 0, 0, 1, ...


Top 50 recent answers are included