lcrmorin
  • Member for 7 years, 6 months
  • Last seen this week
  • France
2 answers
7 votes
236 views
Duplicated features for gradient descent
Accepted answer
7 votes

In 'Efficient Backprop' by Lecun and others (http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf), they explain why correlated variables are bad (§ 4.3 normalizing the inputs). Duplicated data is a ...

View answer
4 answers
5 votes
200 views
Is an $F_1$ score of 0.1 always bad?
Accepted answer
5 votes

From a credit scoring point of view : a $F_1$ score of $0.1$ seems pretty bad but not impossible with an unbalanced data-set. It might be enough for your needs (once you weight your errors by the cost)...

View answer
1 answers
4 votes
67 views
What are the model parameters in PCA?
Accepted answer
3 votes

From an algebra point of view PCA is a base change. You can write the transformation as: $$ T = XW $$ Where $X$ is an nxp matrix (n instance, p features) and $W$ is a pxp 'weight' matrix (whose ...

View answer
2 answers
1 votes
39 views
Unsupervised Learning - Using the Outcome of Learning
3 votes

Unsupervised Learning is a collection of tools and you can use those tools for many purposes. Sometime, it's just for data visualisation / data exploration (see UMAP for example), to better understand ...

View answer
3 answers
1 votes
55 views
Time series classification, without the time dimension
3 votes

Honestly it seems you are quite far from what would need a supervised vision approach. I suggest you to try a simple non-ML approach first : extract text with a standard library then just label what ...

View answer
1 answers
2 votes
285 views
Scikit-learn SelectKBest is picking up obviously unwanted Features
Accepted answer
3 votes

There seems to be two possible approaches to your problem : If they are just identification features that you know aren't informative, you should remove them yourself. SelectKBest - like almost any ...

View answer
3 answers
27 votes
97k views
How to get p-value and confident interval in LogisticRegression with sklearn?
3 votes

This is still not implemented and not planned as it seems out of scope of sklearn, as per Github discussion #6773 and #13048. However, the documentation on linear models now mention that (P-value ...

View answer
4 answers
3 votes
732 views
Is it good practice to convert columns with a number to a range between 0 and 1?
3 votes

I think there si some confusion with the quantile transformer : https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.QuantileTransformer.html#sklearn.preprocessing....

View answer
2 answers
17 votes
63k views
How to adjust the hyperparameters of MLP classifier to get more perfect performance
3 votes

As a complement to the very practical answer of @BrunoGL, I'd like to give a more theoretical answer. I'd like to suggest everyone trying to adjust hyperparameters of a simple Neural Network to read ...

View answer
2 answers
4 votes
601 views
Compare Coefficients of Different Regression Models
3 votes

It's not actually possible to directly compare model coefficients. What you might do that would make more sense is to compare similar metrics. A good start would be to learn about explainability ...

View answer
3 answers
9 votes
15k views
what is darknet and why is it needed for YOLO object detection?
3 votes

https://pjreddie.com/darknet/ is their website... I cite : "Darknet: Open Source Neural Networks in C Darknet is an open source neural network framework written in C and CUDA. It is fast, easy to ...

View answer
1 answers
4 votes
47 views
Sub-sampling so that sample statistics match population statistics
Accepted answer
3 votes

The term you are looking for is stratified sampling : https://en.wikipedia.org/wiki/Stratified_sampling. It's a way to sample from population that can be partitioned into sub-populations. More ...

View answer
4 answers
14 votes
22k views
Is feature engineering still useful when using XGBoost?
2 votes

An empirical answer to that question woud be to look at public kaggle competitions / notebooks (see here), where xgboost is heavily used as state of the art for tabular data problems. The answer is ...

View answer
1 answers
2 votes
144 views
Why is it okay to set the bias vector up with zeros, and not the weight matrices?
2 votes

As per Efficient Backprop from Lecun (§4.6) weight should be initialized in the linear region of the activation function. If they are too big, activation function will saturate and provide small ...

View answer
1 answers
4 votes
487 views
Is it valid to compare SHAP values across models?
2 votes

Shapley values were designed in the context of game theory (source), to share value created by a coalition of player in a game. It has multiple properties, including linearity. The linearity ensure ...

View answer
1 answers
1 votes
71 views
Why don't we find the analytical function of the cost function?
2 votes

The power of networks come from hidden layers with non-linear activation functions. Said non-linear activation function make the calculation of an analytical solution impossible (except maybe for some ...

View answer
2 answers
1 votes
107 views
How to build a model on a dataset having 40% missing values in most of the variables?
2 votes

Generally speaking, you should investigate the process by which your values are missing and try to deal with it. I assume you checked that : There is no meaningfull way to fill those missing values. ...

View answer
1 answers
2 votes
66 views
How curvature information in second order optimization methods helps
Accepted answer
2 votes

You need to consider two steps of your first order optimisation process to see why a second order method can be usefull. (For more clarity we'll work in one dimension). First step : calculate the ...

View answer
1 answers
1 votes
1k views
xgboost classifier predicted negative probabilities
Accepted answer
2 votes

You have to set the option objective = binary:logistic to get probabilities between 0 and 1, otherwise you only get relative scores.

View answer
1 answers
5 votes
184 views
How do I interpret my result of clustering?
2 votes

K-means don't modify the underlying structure of your data. K-means will just provide the 'color' part of your graph. To answer the question about why do you get a cuboid, it's because your ...

View answer
3 answers
3 votes
122 views
Constructing function - f(x,y) for the given minimums (Python)
Accepted answer
2 votes

One way to go is to define functions that are non-zero only near your 'minimal' points, and add them. The goal is to avoid overlap, such that a function associated with a point won't modify the value ...

View answer
1 answers
3 votes
780 views
For outliers treatment, clipping, winsorizing or removing?
Accepted answer
2 votes

On the difference between winsorizing and clipping : The techniques are very similar. They deal with extreme values (that are not necessarily outliers). Imo you should generally avoid thinking that ...

View answer
6 answers
9 votes
3k views
Is it possible to cluster data according to a target?
1 votes

One approach I would try would be a supervised dimension reduction (UMAP for example https://umap-learn.readthedocs.io/en/latest/supervised.html) then a clustering approach (such as Hdbscan: https://...

View answer
1 answers
0 votes
30 views
Sub-word tokenization preprocessing to use transformer
Accepted answer
1 votes

This look like a tqdm problem. Both the module tqdm and the main function tqdm have the same name. This often create some problem as people will just: import tqdm When the right import is: from tqdm ...

View answer
2 answers
0 votes
44 views
Different results after each training of Keras/TensorFlow model
1 votes

It’s not wrong. NN training is inherently stochastic. As an optimisation problem, the tuning of a NN depends on the initialisation (initialisation of the weights). So the result (the local minimum you ...

View answer
3 answers
0 votes
77 views
Confusion Matrix before and after SMOTE is same
1 votes

From your confusion matrix, your model only predict Benign class. It seems that you have a degenerate model. It means that there is a problem somewhere, either your model hasn't learned at all or it ...

View answer
2 answers
3 votes
358 views
High Recall but too low Precision result in imbalanced data
1 votes

It is important to understand that your precision and recall are associated with a binary decision threshold. Basically, the outputs of the model are converted to a binary decision using this ...

View answer
1 answers
0 votes
37 views
PCA ? after the transformed data, are they still same with original data, (if maintain same dimensional)
Accepted answer
1 votes

As you can see from your pics there is a change. From an algebra point of view, PCA is a base change. The interest of the algo is how the target base is designed: it is designed as to explain variance ...

View answer
1 answers
1 votes
17 views
How many run should we implement on the machine learning model?
1 votes

It depends on your problem, start with a big number (32 ?) then decrease if the model is stable / when you deal with stability. You need to know that having multiple runs may be part of the solution: ...

View answer
2 answers
1 votes
49 views
Is there a good systematic approach to explore and analyze data (prior to modelling)?
Accepted answer
1 votes

I assume you are asking about tabular data not vision or NLP. It doesn't exist because 1) there is so much type of data and weird problems and 2) univariate EDA is generally not enough. I can detail ...

View answer
1
2 3 4 5