Imran
  • Member for 6 years
  • Last seen more than a month ago
Updating the weights of the filters in a CNN
Accepted answer
15 votes

In a normal neural network, each neuron has its own weight. This is not correct. Every connection between neurons has its own weight. In a fully connected network each neuron will be associated with ...

View answer
How to get predicted class labels in convolution neural network?
Accepted answer
14 votes

What you have are predicted class probabilities. Since you are doing binary classification, each output is the probability of the first class for that test example. To convert these to class labels ...

View answer
So what's the catch with LSTM?
12 votes

You are right that LSTMs work very well for some problems, but some of the drawbacks are: LSTMs take longer to train LSTMs require more memory to train LSTMs are easy to overfit Dropout is much ...

View answer
Is there a thumb-rule for designing neural-networks?
Accepted answer
11 votes

This question has been answered in detail on CrossValidated: How to choose the number of hidden layers and nodes in a feedforward neural network? However, let me add my own two cents: There is no ...

View answer
XGboost - Choice made by model
9 votes

You can use the ELI5 library to explain the feature contributions to individual predictions for XGBoost models. See Explaining Predictions in the docs, copied below: To get a better idea of how ...

View answer
Is reseating passengers a reinforcement learning problem?
7 votes

Reinforcement learning is more about interacting with an environment, and while this could be posed as an RL problem, I think using Global Optimization would be a more direct approach. Essentially ...

View answer
Sliding window leads to overfitting in LSTM?
6 votes

LSTMs do not require a sliding window of inputs. They can remember what they have seen in the past, and if you feed in training examples one at a time they will choose the right size window of inputs ...

View answer
How to import image data into python for keras?
Accepted answer
6 votes

The docs for ImageDataGenerator suggest that no augmentation is done by default. So you could instantiate it without any augmentation parameters and keep the rest of your code for handling your ...

View answer
Does it make sense that datetime encodes one-hot-vector like one-hot-encoding or something else like
6 votes

If you are trying to predict future values then it doesn't make sense to treat them as categorical features. There is nothing you will learn that can predict future data, since you won't see those ...

View answer
Filter row depending on specific object value and delete those instances
Accepted answer
5 votes

You can replace ? with nan and use dropna(). This will work if you don't already have rows with nan entries that you want to keep. train = train.replace('?', np.nan).dropna() Another option is to ...

View answer
Auto-Encoder to condense (pre-process) large one-hot input vectors?
4 votes

Leland's answer is exactly correct regarding why an autoencoder wouldn't be useful. Let me expand upon that point: Autoencoders and other dimensionality reduction techniques attempt to keep objects ...

View answer
Classification or regression? Which model is more accurate if I only care about being above or under the threshold?
Accepted answer
4 votes

Classification is the more direct approach and it will likely give better results. This is because the model's goal is exactly the same as your goal - i.e. predicting whether the price is above or ...

View answer
Multi target classification for different types of target variables
4 votes

You have one classification task and one regression task, but sklearn's multioutput meta-estimators only support two tasks of the same type. The best solution here is to train two models: A binary ...

View answer
Neural Network with Connections to all forward layers
Accepted answer
4 votes

What you describe has been explored in Deep Residual Neural Networks. A residual block will combine two or more blocks from a standard architecture like a CNN with a skip connection that adds the ...

View answer
How does sklearn KNeighborsClassifier compute class probabilites?
Accepted answer
4 votes

The class probabilities are the normalized weighted average of indicators for the k-nearest classes, weighted by the inverse distance. For example: Say we have 6 classes, and the 5 nearest examples ...

View answer
How to feed my JSON dataset in Keras for character level text classification
3 votes

As mentioned by @Frankstr you want a tutorial on "character-level classification", not handwritten digit recognition as the one you have linked. Character-level classification is typically done with ...

View answer
Non Deterministc Dimensionality reduction
Accepted answer
3 votes

Autoencoders are non-deterministic, since they rely on a random weight initialization.

View answer
Which algorithm to apply for choosing the right point
3 votes

Here are a few ways you might use neural networks to solve this problem: With a plain Feedforward Neural Network: Scale your data to fit in the square around the origin from (-1,-1) to (1,1) ...

View answer
When are weights updated in CNN?
3 votes

Back-propagation technically refers to computing the gradient of the loss function with respect to the parameters. According to Section 6.5 of the Deep Learning book: The term back-propagation is ...

View answer
how should I measure performance if there is no test data?
3 votes

If you are working with only enough data for training and validation, consider using K-Fold Cross Validation: One of the main reasons for using cross-validation instead of using the conventional ...

View answer
Poker tournament winner prediction
Accepted answer
2 votes

As oW_ mentioned, this can be done with the Elo rating system. More specifically, given a record of wins and losses for some player pool, you can fit ratings for each player such the expected result ...

View answer
How does action get selected in a Policy Gradient Method?
Accepted answer
2 votes

You are on the right track. We no longer select an action that we think maximizes the score. Rather we predict what the best action to take is. This can be very effective in large or continuous state ...

View answer
Is the multilayer perceptron only able to accept 1d vector of input data? If yes, why is this so?
2 votes

You could say every type of neural network gets 1d input data. It's just more convenient to think about 2d-CNNs taking 2d data because the convolution operation is best illustrated by moving squares ...

View answer
Changing multiple models into 1 model
Accepted answer
2 votes

This questions is best posed in its original domain, ie medicine, because it will require significant domain knowledge about the problem structure and the nature of the data to reason about how well a ...

View answer
Algorithm for multiple input single output ML
Accepted answer
2 votes

Since you believe the output can be predicted by a linear combination of the inputs, a reasonable approach to try is Linear Regression, specifically Multiple Regression since you have more than one ...

View answer
Neural Network for Multiple Float Output
2 votes

Just use an output layer with 12 neurons instead of 1. Qualitatively there is no difference. For regression the output activations should be linear, and you have a few choices for cost: RMSE, MAE, or ...

View answer
Mean and Variance of Feature Scaling
2 votes

It doesn't make sense to standardize your test set with the mean and variance computed on the test set. At best the test mean and variance will be close enough to the training mean and variance to not ...

View answer
Classes of neural nets and their applications
Accepted answer
2 votes

Convolutional Neural Networks have consistently outperformed other methods for image recognition and related tasks. In fact, beyond a certain image size it is not practical to train a fully-connected ...

View answer
Regression and Neural networks
Accepted answer
2 votes

The problem is that your error is accumulating and diverging. In other words, a small error in the first prediction is leading to a larger error in your second prediction, which is leading to an even ...

View answer
What do you call a feature that always has the same value?
1 votes

After some more research, I believe this is a type of "redundant" feature, in the language of Machine Learning. A redundant feature is one that can be proven to add no information by looking at the ...

View answer