Trjtdtk

Has any country ever had 2 former presidents in jail simultaneously?

Taxes on Dividends in a Roth IRA

What kind of floor tile is this?

Why do Radio Buttons not fill the entire outer circle?

Is there any evidence that Cleopatra and Caesarion considered fleeing to India to escape the Romans?

What is the difference between lands and mana?

Shouldn’t conservatives embrace universal basic income?

Has the laser at Magurele, Romania reached a tenth of the Sun's power?

15% tax on $7.5k earnings. Is that right?

awk assign to multiple variables at once

Circuit Analysis: Obtaining Close Loop OP - AMP Transfer function

Why is the Sun approximated as a black body at ~ 5800 K?

Why Shazam when there is already Superman?

Delete multiple columns using awk or sed

Will number of steps recorded on FitBit/any fitness tracker add up distance in PokemonGo?

Why should universal income be universal?

Are Captain Marvel's powers affected by Thanos breaking the Tesseract and claiming the stone?

What does "Scientists rise up against statistical significance" mean? (Comment in Nature)

Is a Java collection guaranteed to be in a valid, usable state after a ConcurrentModificationException?

Is this part of the description of the Archfey warlock's Misty Escape feature redundant?

Microchip documentation does not label CAN buss pins on micro controller pinout diagram

Why do ¬, ∀ and ∃ have the same precedence?

When were female captains banned from Starfleet?

Giving feedback to someone without sounding prejudiced

How to optimize NN seq2seq classification training with time-dependent class imbalance?

Time dependent Classification problemclassification algorithms for anomaly detection when there is class imbalanceHow to overcome the class imbalance issue in a text classification problem?Handling data imbalance and class number for classificationDealing with a dataset where a subset of points live in a higher dimensional spaceTraining neural network classifier with one class after anotherhow to optimize the weights of a neural net when feeding it with multiple training samples?Image classification with different number of training image for each classMultivariate time Series classification - One classIs it possible to have too few labels to train a model?

0

$begingroup$

I want to classify timeseries using seq2seq since they have a high chance of being "corrupted" towards the end.

The traces are labelled e.g.

111111100000
555500000000
330000000000
444444440000
444444444444

as you can see, they (almost) always end up in zero-land.

If I plot the label-distribution over time, it looks something like the plot below, where the trace has an increasingly growing chance of being terminated and zero-ed at some point.

Are there any common methods for balancing problems like this, or should I simply try to balance classes based on e.g. the first n datapoints only?

neural-network deep-learning classification sequence-to-sequence

share|improve this question

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

$endgroup$

add a comment | 

0

$begingroup$

I want to classify timeseries using seq2seq since they have a high chance of being "corrupted" towards the end.

The traces are labelled e.g.

111111100000
555500000000
330000000000
444444440000
444444444444

as you can see, they (almost) always end up in zero-land.

If I plot the label-distribution over time, it looks something like the plot below, where the trace has an increasingly growing chance of being terminated and zero-ed at some point.

Are there any common methods for balancing problems like this, or should I simply try to balance classes based on e.g. the first n datapoints only?

neural-network deep-learning classification sequence-to-sequence

share|improve this question

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

$endgroup$

add a comment | 

$begingroup$

I want to classify timeseries using seq2seq since they have a high chance of being "corrupted" towards the end.

The traces are labelled e.g.

111111100000
555500000000
330000000000
444444440000
444444444444

as you can see, they (almost) always end up in zero-land.

If I plot the label-distribution over time, it looks something like the plot below, where the trace has an increasingly growing chance of being terminated and zero-ed at some point.

Are there any common methods for balancing problems like this, or should I simply try to balance classes based on e.g. the first n datapoints only?

neural-network deep-learning classification sequence-to-sequence

share|improve this question

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

$endgroup$

111111100000
555500000000
330000000000
444444440000
444444444444

share|improve this question

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

share|improve this question

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114

asked Mar 18 at 17:45

komodovaran_komodovaran_

1114