I am new to XGBoost and I am currently working on a project where we have built an XGBoost classifier. Now we want to run some feature selection techniques. Is backward elimination method a good idea for this? I have used it in regression but I am not sure if/how to use it in a classification problem. Any leads will be greatly appreciated.
Note: I have already tried permutation line importance and it has yielded good results! Looking for another method to evaluate the features in the model.
Consider asking your question on Cross Validated since feature selection is more about theory/practice than code.
What is your concern ? Remove "noisy" features who drive down your results, obtain a sparse model ? Backward selection is one way to do of course. That being said, not sure if you are aware of this but XGBoost computes its own "variable importance" values.
# plot feature importance using built-in function
from xgboost import XGBClassifier
from xgboost import plot_importance
from matplotlib import pyplot
model = XGBClassifier()
model.fit(X, y)
# plot feature importance
plot_importance(model)
pyplot.show()
Something like this. This importance is based on how many times a feature is used to make a split. You can then define for instance a threshold below which you do not keep the variables. However do not forget that :
This variable importance has been obtained on the training data only
The removal of a variable with high importance may not affect your prediction error, e.g. if it is correlated with another highly important variable. Other tricks such as this one may exist.
Related
I am not an expert on logistic regression, but I thought when solving it using lgfgs it was doing optimization, finding local minima for the objective function. But every time I run it using scikit-learn, it is returning the same results, even when I feed it a different random state.
Below is code that reproduces my issue.
First set up the problem by generating data
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn import metrics
from sklearn import datasets
# generate data
X, y = datasets.make_classification(n_samples=1000,
n_features=10,
n_redundant=4,
n_clusters_per_class=1,
random_state=42)
# Set up the test/training data
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.25)
Second, train the model and inspect results
# Set up a different random state each time
rand_state = np.random.randint(1000)
print(rand_state)
model = LogisticRegression(max_iter=1000,
solver='lbfgs',
random_state=rand_state)
model.fit(X_train,y_train)
y_pred = model.predict(X_test)
conf_mat = metrics.confusion_matrix(y_test, y_pred)
print(y_pred[:20],"\n", conf_mat)
I get the same y_pred (and obviously confusion matrix) every time I run this even though I'm using the lbfgs solver with a different random state each run. I'm confused, as I thought this was a stochastic solver that was traveling down a gradient into a local minimum.
Maybe I'm not properly randomizing the initial state? I haven't been able to figure it out from the documentation.
Discussion of Related Question
There is a related question, which I didn't find during my research:
Does logistic regression always find global optimum, assuming that the optimisation converges?
The answer there is that the cost function is convex, so if the numerical solution is well-behaved, it will find a global minimum. That is, there aren't a bunch of local minima that your optimization algorithm will get stuck in: it will reach the same (global) minimum each time (perhaps depending on the solver you choose?).
However, in the comments someone pointed out, depending on what solvers you choose there are cases when you will not reach the same solution, that it depends on the random_state parameter. At the very least, I think this would be helpful to resolve.
First, let me put in the answer what got this closed as duplicate earlier: a logistic regression problem (without perfect separation) has a global optimum, and so there are no local optima to get stuck in with different random seeds. If the solver converges satisfactorily, it will do so on the global optimum. So the only time random_state can have any effect is when the solver fails to converge.
Now, the documentation for LogisticRegression's parameter random_state states:
Used when solver == ‘sag’, ‘saga’ or ‘liblinear’ to shuffle the data. [...]
So for your code, with solver='lbfgs', indeed there is no expected effect.
It's not too hard to make sag and saga fail to converge, and with different random_states to end at different solutions; to make it easier, set max_iter=1. liblinear apparently does not use the random_state unless solving the dual, so also setting dual=True admits different solutions. I found that thanks to this comment on a github issue (the rest of the issue may be worth reading for more background).
I have a business problem, I have run the regression model in python to predict my target value. When validating it with my test set I came to know that my predicted variable is very far from my actual value. Now the thing I want to extract from this model is that, which feature played the role to deviate my predicted value from actual value (let say difference is in some threshold value)?
I want to rank the features impact wise so that I could address to my client.
Thanks
It depends on the estimator you chose, linear models often have a coef_ method you can call to get the coef used for each feature, given they are normalized this tells you what you want to know.
As told above for tree model you have the feature importance. You can also use libraries like treeinterpreter described here:
Interpreting Random Forest
examples
You can have a look at this -
Feature selection
Check the Random Forest Regressor - for performing Regression.
# Example
from sklearn.ensemble import RandomForestRegressor
from sklearn.datasets import make_regression
X, y = make_regression(n_features=4, n_informative=2,
random_state=0, shuffle=False)
regr = RandomForestRegressor(max_depth=2, random_state=0,
n_estimators=100)
regr.fit(X, y)
print(regr.feature_importances_)
print(regr.predict([[0, 0, 0, 0]]))
Check regr.feature_importances_ for getting the higher, more important features. Further information on FeatureImportance
Edit-1:
As pointed out in user (#blacksite) comment, only feature_importance does not provide complete interpretation of Random forest. For further analysis of results and responsible Features. Please refer to following blogs
https://medium.com/usf-msds/intuitive-interpretation-of-random-forest-2238687cae45 (preferred as it provides multiple techniques )
https://blog.datadive.net/interpreting-random-forests/ (focuses on 1 technique but also provides python library - treeinterpreter)
More on feature_importance:
If you simply use the feature_importances_ attribute to select the
features with the highest importance score. Feature selection using
feature
importances
Feature importance also depends on the criteria used for splitting
and calculating importance Interpreting Decision Tree in context of
feature
importances
I want to build a Random Forest Regressor to model count data (Poisson distribution). The default 'mse' loss function is not suited to this problem. Is there a way to define a custom loss function and pass it to the random forest regressor in Python (Sklearn, etc..)?
Is there any implementation to fit count data in Python in any packages?
In sklearn this is currently not supported. See discussion in the corresponding issue here, or this for another class, where they discuss reasons for that a bit more in detail (mainly the large computational overhead for calling a Python function).
So it could be done as discussed within the issues, by forking sklearn, implementing the cost function in Cython and then adding it to the list of available 'criterion'.
If the problem is that the counts c_i arise from different exposure times t_i, then indeed one cannot fit the counts, but one can still fit the rates r_i = c_i/t_i using MSE loss function, where one should, however, use weights proportional to the exposures, w_i = t_i.
For a true Random Forest Poisson regression, I've seen that in R there is the rpart library for building a single CART tree, which has a Poisson regression option. I wish this kind of algorithm would have been imported to scikit-learn.
In R, writing a custom objective function is fairly simple.
randomForestSRC package in R has provision for writing your own custom split rule. The custom split rule, however has to be written in pure C language.
All you have to do is, write your own custom split rule, register the split rule, compile and install the package.
The custom split rule has to be defined in the file called splitCustom.c in randomForestSRC source code.
You can find more info
here.
The file in which you define the split rule is
this.
scikit-learn has two logistic regression functions:
sklearn.linear_model.LogisticRegression
sklearn.linear_model.LogisticRegressionCV
I'm just curious what the CV stands for in the second one. The only acronym I know in ML that matches "CV" is cross-validation, but I'm guessing that's not it, since that would be achieved in scikit-learn with a wrapper function, not as part of the logistic regression function itself (I think).
You are right in guessing that the latter allows the user to perform cross validation. The user can pass the number of folds as an argument cv of the function to perform k-fold cross-validation (default is 10 folds with StratifiedKFold).
I would recommend reading the documentation for the functions LogisticRegression and LogisticRegressionCV
Yes, it's cross-validation. Excerpt from the docs:
For the grid of Cs values (that are set by default to be ten values in a logarithmic scale between 1e-4 and 1e4), the best hyperparameter is selected by the cross-validator StratifiedKFold, but it can be changed using the cv parameter.
The point here is the following:
yes: sklearn has general model-selection wrappers providing CV-functionality for all those classifiers/regressors
but: when the classifier/regressor is known/fixed a-priori (to some extent) or sometimes even some CV-model, one can gain advantages using these facts with specialized code bound to one classifier/regressor resulting in improved performance!
Typically:
CV already embedded in optimization-algorithm
Efficient warm-starting (instead of full re-optimization after just the change of one parameter like alpha)
It seems, at least the latter idea is used in sklearn's LogisticRegressionCV, as seen in this excerpt:
In the case of newton-cg and lbfgs solvers, we warm start along the path i.e guess the initial coefficients of the present fit to be the coefficients got after convergence in the previous fit, so it is supposed to be faster for high-dimensional dense data.
May I also refer you to this section in scikit-learn documentation which I beleive explains it well:
Some models can fit data for a range of values of some parameter
almost as efficiently as fitting the estimator for a single value of
the parameter. This feature can be leveraged to perform a more
efficient cross-validation used for model selection of this parameter.
The most common parameter amenable to this strategy is the parameter
encoding the strength of the regularizer. In this case we say that we
compute the regularization path of the estimator.
And logistic regression is one such model. That's why scikit-learn has the dedicated LogisticRegressionCV class that does this.
There are some things left out on other answers, e.g. about gridsearch functionality. See the docs:
cross-validation estimator
An estimator that has built-in cross-validation capabilities to automatically select the best hyper-parameters (see the User Guide). Some example of cross-validation estimators are ElasticNetCV and LogisticRegressionCV. Cross-validation estimators are named EstimatorCV and tend to be roughly equivalent to GridSearchCV(Estimator(), ...). The advantage of using a cross-validation estimator over the canonical estimator class along with grid search is that they can take advantage of warm-starting by reusing precomputed results in the previous steps of the cross-validation process. This generally leads to speed improvements. An exception is the RidgeCV class, which can instead perform efficient Leave-One-Out CV.
https://scikit-learn.org/stable/glossary.html#term-cross-validation-estimator
https://github.com/amueller/talks_odt/blob/master/2015/nyc-open-data-2015-andvanced-sklearn.pdf
I have a data set containing 1000 points each with 2 inputs and 1 output. It has been split into 80% for training and 20% for testing purpose. I am training it using sklearn support vector regressor. I have got 100% accuracy with training set but results obtained with test set are not good. I think it may be because of overfitting. Please can you suggest me something to solve the problem.
You may be right: if your model scores very high on the training data, but it does poorly on the test data, it is usually a symptom of overfitting. You need to retrain your model under a different situation. I assume you are using train_test_split provided in sklearn, or a similar mechanism which guarantees that your split is fair and random. So, you will need to tweak the hyperparameters of SVR and create several models and see which one does best on your test data.
If you look at the SVR documentation, you will see that it can be initiated using several input parameters, each of which could be set to a number of different values. For the simplicity, let's assume you are only dealing with two parameters that you want to tweak: 'kernel' and 'C', while keeping the third parameter 'degree' set to 4. You are considering 'rbf' and 'linear' for kernel, and 0.1, 1, 10 for C. A simple solution is this:
for kernel in ('rbf', 'linear'):
for c in (0.1, 1, 10):
svr = SVR(kernel=kernel, C=c, degree=4)
svr.fit(train_features, train_target)
score = svr.score(test_features, test_target)
print kernel, c, score
This way, you can generate 6 models and see which parameters lead to the best score, which will be the best model to choose, given these parameters.
A simpler way is to let sklearn to do most of this work for you, using GridSearchCV (or RandomizedSearchCV):
parameters = {'kernel':('linear', 'rbf'), 'C':(0.1, 1, 10)}
clf = GridSearchCV(SVC(degree=4), parameters)
clf.fit(train_features, train_target)
print clf.best_score_
print clf.best_params_
model = clf.best_estimator_ # This is your model
I am working on a little tool to simplify using sklearn for small projects, and make it a matter of configuring a yaml file, and letting the tool do all the work for you. It is available on my github account. You might want to take a look and see if it helps.
Finally, your data may not be linear. In that case you may want to try using something like PolynomialFeatures to generate new nonlinear features based on the existing ones and see if it improves your model quality.
Try fitting your data using training data split Sklearn K-Fold cross-validation, this provides you a fair split of data and better model , though at a cost of performance , which should really matter for small dataset and where the priority is accuracy.
A few hints:
Since you have only two inputs, it would be great if you plot your data. Try either a scatter with alpha = 0.3 or a heatmap.
Try GridSearchCV, as mentioned by #shahins.
Especially, try different values for the C parameter. As mentioned in the docs, if you have a lot of noisy observations you should decrease it. It corresponds to regularize more the estimation.
If it's taking too long, you can also try RandomizedSearchCV
As a side note from #shahins answer (I am not allowed to add comments), both implementations are not equivalent. GridSearchCV is better since it performs cross-validation in the training set for tuning the hyperparameters. Do not use the test set for tuning hyperparameters!
Don't forget to scale your data