Both lightgbm and sklearn's HistGradientBoostingClassifier estimators use histograms to decide on best splits for continuous features.
Is it possible to explain intuitively (or with some example) the process of histogram creation and how does it help in deciding in faster split point at a node.
I have looked for answers extensively over the Internet but could not find any simple or intuitive way as to how histograms are constructed.
I am not sure but it could be related to how (unique) Regression trees are constructed in XGBoost. For a continuous feature, you construct an histogram, decide on the split (e.g. weight < 70kg), construct a Regression tree and compute the Similarity score as well as the Gain. However, when the range of the values in the continuous feature is quite large then it is quite computationally expensive to try all the possible split values. In that case, XGBoost basically makes the split by making use of the quantiles which involves dividing all the observations into equally sized sets.
I guess sklearn's HistGradientBoostingClassifier might involve the above tool optimization as well for coming up with the best split.
Related
I want to use a clustering algorithm to a dataframe that contains a lot of features (32 columns).
A part of the features are encoded using one hot encoder.
I want to use PCA ( Principal Component analysis ) to reduce the dimension and make the machine learning process easier.
Is it possible to use the PCA just for some columns of the data frame and keep the other columns as they are then use machine learning model.
Or it is obligatory to use PCA for all the dataframe before clustering.
I guess there should be no issue with doing what you describe.
What this does, effectively, is merge some of the objects' features into fewer ones, but then using other, non-merged ones in addition to the merged ones. I don't know what effect that would have on the outcome; it might be good to run a correlation to see whether the unmerged features add anything to the PCA-merged ones. You might find that they basically duplicate what is there already.
Since clustering is an exploratory method, you can basically do whatever you want. It is of course advisable to have a reason for doing so, as it otherwise ends up as simply trial-and-error, and if you find a result, you won't be able to describe why you got there. It is possible (or even likely for some data sets) that there are multiple ways to cluster them, so you should make decisions based on what you know about the data already, so they can be justified in those terms.
Running random trial-and-error clustering until you find a structure makes it a bit difficult to come up with a good explanation why that structure is valid.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 2 years ago.
Improve this question
I am trying to tune a basic neural network as practice. (Based on an example from a coursera course: Neural Networks and Deep Learning - DeepLearning.AI)
I face the issue of the random weight initialization. Lets say I try to tune the number of layers in the network.
I have two options:
1.: set the random seed to a fixed value
2.: run my experiments more times without setting the seed
Both version has pros and cons.
My biggest concern is that if I use a random seed (e.g.: tf.random.set_seed(1)) then the determined values can be "over-fitted" to the seed and may not work well without the seed or if the value is changed (e.g.: tf.random.set_seed(1) -> tf.random.set_seed(2). On the other hand, if I run my experiments more times without random seed then I can inspect less option (due to limited computing capacity) and still only inspect a subset of possible random weight initialization.
In both cases I feel that luck is a strong factor in the process.
Is there a best practice how to handle this topic?
Has TensorFlow built in tools for this purpose? I appreciate any source of descriptions or tutorials. Thanks in advance!
Tuning hyperparameters in deep learning (generally in machine learning) is a common issue. Setting the random seed to a fixed number ensures reproducibility and fair comparison. Repeating the same experiment will lead to the same outcomes. As you probably know, best practice to avoid over-fitting is to do a train-test split of your data and then use k-fold cross-validation to select optimal hyperparameters. If you test multiple values for a hyperparameter, you want to make sure other circumstances that might influence the performance of your model (e.g. train-test-split or weight initialization) are the same for each hyperparameter in order to have a fair comparison of the performance. Therefore I would always recommend to fix the seed.
Now, the problem with this is, as you already pointed out, the performance for each model will still depend on the random seed, like the particular data split or weight initialization in your case. To avoid this, one can do repeated k-fold-cross validation. That means you repeat the k-fold cross-validation multiple times, each time with a different seed, select best parameters of that run, test on test data and average the final results to get a good estimate of performance + variance and therefore eliminate the influence the seed has in the validation process.
Alternatively you can perform k-fold cross validation a single time and train each split n-times with a different random seed (eliminating the effect of weight initialization, but still having the effect of the train-test-split).
Finally TensorFlow has no build-in tool for this purpose. You as practitioner have to take care of this.
There is no an absolute right or wrong answer to your question. You are almost answered your own question already. In what follows, however, I will try to expand more, via the following points:
The purpose of random initialization is to break the symmetry that makes neural networks fail to learn:
... the only property known with complete certainty is that the
initial parameters need to “break symmetry” between different units.
If two hidden units with the same activation function are connected to
the same inputs, then these units must have different initial
parameters. If they have the same initial parameters, then a
deterministic learning algorithm applied to a deterministic cost and
model will constantly update both of these units in the same way...
Deep Learning (Adaptive Computation and Machine Learning series)
Hence, we need the neural network components (especially weights) to be initialized by different values. There are some rules of thumb of how to choose those values, such as the Xavier initialization, which samples from normal distribution with mean of 0 and special variance based on the number of the network layer. This is a very interesting article to read.
Having said so, the initial values are important but not extremely critical "if" proper rules are followed, as per mentioned in point 2. They are important because large or improper ones may lead to vanishing or exploding gradient problems. On the other hand, different "proper" weights shall not hugely change the final results, unless they are making the aforementioned problems, or getting the neural network stuck at some local maxima. Please note, however, the the latter depends also on many other aspects, such as the learning rate, the activation functions used (some explode/vanish more than others: this is a great comparison), the architecture of the neural network (e.g. fully connected, convolutional ..etc: this is a cool paper) and the optimizer.
In addition to point 2, bringing a good learning optimizer into the bargain, other than the standard stochastic one, shall in theory not let a huge influence of the initial values to affect the final results quality, noticeably. A good example is Adam, which provides a very adaptive learning technique.
If you still get a noticeably-different results, with different "proper" initialized weights, there are some ways that "might help" to make neural network more stable, for example: use a Train-Test split, use a GridSearchCV for best parameters, and use k-fold cross validation...etc.
At the end, obviously the best scenario is to train the same network with different random initial weights many times then get the average results and variance, for more specific judgement on the overall performance. How many times? Well, if can do it hundreds of times, it will be better, yet that clearly is almost impractical (unless you have some Googlish hardware capability and capacity). As a result, we come to the same conclusion that you had in your question: There should be a tradeoff between time & space complexity and reliability on using a seed, taking into considerations some of the rules of thumb mentioned in previous points. Personally, I am okay to use the seed because I believe that, "It’s not who has the best algorithm that wins. It’s who has the most data". (Banko and Brill, 2001). Hence, using a seed with enough (define enough: it is subjective, but the more the better) data samples, shall not cause any concerns.
I have a question regarding boxcox transformation(or log transformation). I am working on a data-set which I have lots of skewed features. Now when I take the boxcox transformation, I get quite a nice distribution but the thing is correlation decrease. Now if I was working with linear models I would just consider correlation to decide I should transform the feature or not. But as I mentioned I am working with tree-based models, so should I transform the feature to get a more dispersed distribution or I leave the feature as it is to avoid a decrease in correlation.
I add a screenshot of distribution and its relationship with the target variable, for both transformed and not transformed(Left 2 plots original feature and target).
PS: Guessing from the plots, it seems to me that if I transform the feature it will be easier for tree to find a split for this particular feature.
Thanks a lot,
I have a set of data, 2D matrix (like Grey pictures).
And use CNN for classifier.
Would like to know if there is any study/experience on the accuracy impact
if we change the encoding from traditionnal encoding.
I suppose yes, question is rather which transformation of the encoding make the accuracy invariant, which one deteriorates....
To clarify, this concerns mainly the quantization process of the raw data into input data.
EDIT:
Quantize the raw data into input data is already a pre-processing of the data, adding or removing some features (even minor). It seems not very clear the impact in term of accuracy on this quantization process on real dnn computation.
Maybe, some research available.
I'm not aware of any research specifically dealing with quantization of input data, but you may want to check out some related work on quantization of CNN parameters: http://arxiv.org/pdf/1512.06473v2.pdf. Depending on what your end goal is, the "Q-CNN" approach may be useful for you.
My own experience with using various quantizations of the input data for CNNs has been that there's a heavy dependency between the degree of quantization and the model itself. For example, I've played around with using various interpolation methods to reduce image sizes and reducing the color palette size, and in the end, I discovered that each variant required a different tuning of hyper-parameters to achieve optimal results. Generally, I found that minor quantization of data had a negligible impact, but there was a knee in the curve where throwing away additional information dramatically impacted the achievable accuracy. Unfortunately, I'm not aware of any way to determine what degree of quantization will be optimal without experimentation, and even deciding what's optimal involves a trade-off between efficiency and accuracy which doesn't necessarily have a one-size-fits-all answer.
On a theoretical note, keep in mind that CNNs need to be able to find useful, spatially-local features, so it's probably reasonable to assume that any encoding that disrupts the basic "structure" of the input would have a significantly detrimental effect on the accuracy achievable.
In usual practice -- a discrete classification task in classic implementation -- it will have no effect. However, the critical point is in the initial computations for back-propagation. The classic definition depends only on strict equality of the predicted and "base truth" classes: a simple right/wrong evaluation. Changing the class coding has no effect on whether or not a prediction is equal to the training class.
However, this function can be altered. If you change the code to have something other than a right/wrong scoring, something that depends on the encoding choice, then encoding changes can most definitely have an effect. For instance, if you're rating movies on a 1-5 scale, you likely want 1 vs 5 to contribute a higher loss than 4 vs 5.
Does this reasonably deal with your concerns?
I see now. My answer above is useful ... but not for what you're asking. I had my eye on the classification encoding; you're wondering about the input.
Please note that asking for off-site resources is a classic off-topic question category. I am unaware of any such research -- for what little that is worth.
Obviously, there should be some effect, as you're altering the input data. The effect would be dependent on the particular quantization transformation, as well as the individual application.
I do have some limited-scope observations from general big-data analytics.
In our typical environment, where the data were scattered with some inherent organization within their natural space (F dimensions, where F is the number of features), we often use two simple quantization steps: (1) Scale all feature values to a convenient integer range, such as 0-100; (2) Identify natural micro-clusters, and represent all clustered values (typically no more than 1% of the input) by the cluster's centroid.
This speeds up analytic processing somewhat. Given the fine-grained clustering, it has little effect on the classification output. In fact, it sometimes improves the accuracy minutely, as the clustering provides wider gaps among the data points.
Take with a grain of salt, as this is not the main thrust of our efforts.
I am trying out Haskell to compute partition functions of models in statistical physics. This involves traversing quite large lists of configurations and summing various observables - which I would like to do as efficiently as possible.
The current version of my code is here: https://gist.github.com/2420539
Some strange things happen when trying to choose between lists and vectors to enumerate the configurations; in particular, to truncate the list, using V.toList . V.take (3^n) . V.fromList (where V is Data.Vector) is faster than just using take, which feels a bit counter-intuitive. In both cases the list is evaluated lazily.
The list itself is built using iterate; if instead I use Vectors as much as possible and build the list by using V.iterateN, again it becomes slower ...
My question is, is there a way (other than splicing V.toList and V.fromList at random places in the code) to predict which one will be the quickest? (BTW, I compile everything using ghc -O2 with the current stable version.)
Vectors are strict, and have O(1) subsets (e.g. take). They also have an optimized insert and delete. So you will sometimes see performance improvements by switching data structures on the fly. However, it is usually the wrong approach -- keeping all data in either one form or the other is better. (And you're using UArrays as well -- further confusing the issue).
General rules:
If the data is large and being transformed only in bulk fashion, using a dense, efficient structures like vectors make sense.
If the data is small, and traversed linearly, rarely, then lists make sense.
Remember that operations on lists and vectors have different complexity, so while iterate . replicate on lists is O(n), but lazy, the same on vectors will not necessarily be as efficient (you should prefer the built in methods in vector to generate arrays).
Generally, vectors should always be better for numerical operations. It might be that you have to use different functions that you do in lists.
I would stick to vectors only. Avoid UArrays, and avoid lists except as generators.