I have trained a logistic regression model. Now I have to update(partial fit) the model with new set of training data. Is it possible ?
You cannot use partial_fit on LogisticRegression.
But you can:
use warm_start=True, which reuse the solution of the previous call to fit as initialization, to speed up convergence.
use SGDClassifier with loss='log' which is equivalent to LogisticRegression, and which supports partial_fit.
Note the difference between partial_fit and warm_start. Both methods starts from the previous model and update it, but partial_fit updates only slightly the model, while warm_start goes all the way to convergence on the new training data, forgetting the previous model. warm_start is only used to speed up convergence.
See also the glossary:
warm_start
When fitting an estimator repeatedly on the same dataset, but for multiple parameter values (such as to find the value maximizing performance as in grid search), it may be possible to reuse aspects of the model learnt from the previous parameter value, saving time. When warm_start is true, the existing fitted model attributes an are used to initialise the new model in a subsequent call to fit.
Note that this is only applicable for some models and some parameters, and even some orders of parameter values. For example, warm_start may be used when building random forests to add more trees to the forest (increasing n_estimators) but not to reduce their number.
partial_fit also retains the model between calls, but differs: with warm_start the parameters change and the data is (more-or-less) constant across calls to fit; with partial_fit, the mini-batch of data changes and model parameters stay fixed.
There are cases where you want to use warm_start to fit on different, but closely related data. For example, one may initially fit to a subset of the data, then fine-tune the parameter search on the full dataset. For classification, all data in a sequence of warm_start calls to fit must include samples from each class.
__
partial_fit
Facilitates fitting an estimator in an online fashion. Unlike fit, repeatedly calling partial_fit does not clear the model, but updates it with respect to the data provided. The portion of data provided to partial_fit may be called a mini-batch. Each mini-batch must be of consistent shape, etc.
partial_fit may also be used for out-of-core learning, although usually limited to the case where learning can be performed online, i.e. the model is usable after each partial_fit and there is no separate processing needed to finalize the model. cluster.Birch introduces the convention that calling partial_fit(X) will produce a model that is not finalized, but the model can be finalized by calling partial_fit() i.e. without passing a further mini-batch.
Generally, estimator parameters should not be modified between calls to partial_fit, although partial_fit should validate them as well as the new mini-batch of data. In contrast, warm_start is used to repeatedly fit the same estimator with the same data but varying parameters.
Related
Normalization e.g. z-scoring is a common preprocessing method in Machine Learning.
I am analyzing a dataset and use ensemble methods like Random Forests or the XGBOOST framework.
Now I compare models using
non normalized features
z-scored features
Using crossvalidation I observe in both cases that with higher max_depth parameter the training error decreases.
For the 1. case the test error also decreases and saturates at a certain MAE:
For the z-scored features however the test error is non decreasing at all.
In this question: https://datascience.stackexchange.com/questions/16225/would-you-recommend-feature-normalization-when-using-boosting-trees it was discussed that normalization is not necessary for tree based methods. But the example above shows that it has a severe effect.
So I have two questions regarding this:
Does it imply that overfitting with ensemble based methods is possible even when the test error decreases?
Should normalization like z-scoring always be common practice when working with ensemble methods?
Is it possible that normalization methods decrease the model performance?
Thanks!
It is not easy to see what is going on in the absence of any code or data.
Normalisation may or may not be helpful depending on the particular data and how the normalisation step is applied.
Tree based methods ought to be robust enough to handle the raw data.
In your cross validations is your code doing the normalisation separately for each fold?
Doing a single normalisation prior to cv may lead to significant leakage.
With very high values of depth you will have a much more complex model that will fit the training data well but will fail to generalise to new data.
I tend to prefer max depths from 2 to 5.
If I can't get a reasonable model I turn my efforts to feature engineering rather than trying to tweak the hyperparameters too much.
For a class project, I designed a neural network to approximate sin(x), but ended up with a NN that just memorized my function over the data points I gave it. My NN took in x-values with a batch size of 200. Each x-value was multiplied by 200 different weights, mapping to 200 different neurons in my first layer. My first hidden layer contained 200 neurons, each one a linear combination of the x-values in the batch. My second hidden layer also contained 200 neurons, and my loss function was computed between the 200 neurons in my second layer and the 200 values of sin(x) that the input mapped to.
The problem is, my NN perfectly "approximated" sin(x) with 0 loss, but I know it wouldn't generalize to other data points.
What did I do wrong in designing this neural network, and how can I avoid memorization and instead design my NN's to "learn" about the patterns in my data?
It is same with any machine learning algorithm. You have a dataset based on which you try to learn "the" function f(x), which actually generated the data. In real life datasets, it is impossible to get the original function from the data, and therefore we approximate it using something g(x).
The main goal of any machine learning algorithm is to predict unseen data as best as possible using the function g(x).
Given a dataset D you can always train a model, which will perfectly classify all the datapoints (you can use a hashmap to get 0 error on the train set), but which is overfitting or memorization.
To avoid such things, you yourself have to make sure that the model does not memorise and learns the function. There are a few things which can be done. I am trying to write them down in an informal way (with links).
Train, Validation, Test
If you have large enough dataset, use Train, Validation, Test splits. Split the dataset in three parts. Typically 60%, 20% and 20% for Training, Validation and Test, respectively. (These numbers can vary based on need, also in case of imbalanced data, check how to get stratified partitions which preserve the class ratios in every split). Next, forget about the Test partition, keep it somewhere safe, don't touch it. Your model, will be trained using the Training partition. Once you have trained the model, evaluate the performance of the model using the Validation set. Then select another set of hyper-parameter configuration for your model (eg. number of hidden layer, learaning algorithm, other parameters etc.) and then train the model again, and evaluate based on Validation set. Keep on doing this for several such models. Then select the model, which got you the best validation score.
The role of validation set here is to check what the model has learned. If the model has overfit, then the validation scores will be very bad, and therefore in the above process you will discard those overfit models. But keep in mind, although you did not use the Validation set to train the model, directly, but the Validation set was used indirectly to select the model.
Once you have selected a final model based on Validation set. Now take out your Test set, as if you just got new dataset from real life, which no one has ever seen. The prediction of the model on this Test set will be an indication how well your model has "learned" as it is now trying to predict datapoints which it has never seen (directly or indirectly).
It is key to not go back and tune your model based on the Test score. This is because once you do this, the Test set will start contributing to your mode.
Crossvalidation and bootstrap sampling
On the other hand, if your dataset is small. You can use bootstrap sampling, or k-fold cross-validation. These ideas are similar. For example, for k-fold cross-validation, if k=5, then you split the dataset in 5 parts (also be carefull about stratified sampling). Let's name the parts a,b,c,d,e. Use the partitions [a,b,c,d] to train and get the prediction scores on [e] only. Next, use the partitions [a,b,c,e] and use the prediction scores on [d] only, and continue 5 times, where each time, you keep one partition alone and train the model with the other 4. After this, take an average of these scores. This is indicative of that your model might perform if it sees new data. It is also a good practice to do this multiple times and perform an average. For example, for smaller datasets, perform a 10 time 10-folds cross-validation, which will give a pretty stable score (depending on the dataset) which will be indicative of the prediction performance.
Bootstrap sampling is similar, but you need to sample the same number of datapoints (depends) with replacement from the dataset and use this sample to train. This set will have some datapoints repeated (as it was a sample with replacement). Then use the missing datapoins from the training dataset to evaluate the model. Perform this multiple times and average the performance.
Others
Other ways are to incorporate regularisation techniques in the classifier cost function itself. For example in Support Vector Machines, the cost function enforces conditions such that the decision boundary maintains a "margin" or a gap between two class regions. In neural networks one can also do similar things (although it is not same as in SVM).
In neural network you can use early stopping to stop the training. What this does, is train on the Train dataset, but at each epoch, it evaluates the performance on the Validation dataset. If the model starts to overfit from a specific epoch, then the error for Training dataset will keep on decreasing, but the error of the Validation dataset will start increasing, indicating that your model is overfitting. Based on this one can stop training.
A large dataset from real world tends not to overfit too much (citation needed). Also, if you have too many parameters in your model (to many hidden units and layers), and if the model is unnecessarily complex, it will tend to overfit. A model with lesser pameter will never overfit (though can underfit, if parameters are too low).
In the case of you sin function task, the neural net has to overfit, as it is ... the sin function. These tests can really help debug and experiment with your code.
Another important note, if you try to do a Train, Validation, Test, or k-fold crossvalidation on the data generated by the sin function dataset, then splitting it in the "usual" way will not work as in this case we are dealing with a time-series, and for those cases, one can use techniques mentioned here
First of all, I think it's a great project to approximate sin(x). It would be great if you could share the snippet or some additional details so that we could pin point the exact problem.
However, I think that the problem is that you are overfitting the data hence you are not able to generalize well to other data points.
Few tricks that might work,
Get more training points
Go for regularization
Add a test set so that you know whether you are overfitting or not.
Keep in mind that 0 loss or 100% accuracy is mostly not good on training set.
How train_on_batch() is different from fit()? What are the cases when we should use train_on_batch()?
For this question, it's a simple answer from the primary author:
With fit_generator, you can use a generator for the validation data as
well. In general, I would recommend using fit_generator, but using
train_on_batch works fine too. These methods only exist for the sake of
convenience in different use cases, there is no "correct" method.
train_on_batch allows you to expressly update weights based on a collection of samples you provide, without regard to any fixed batch size. You would use this in cases when that is what you want: to train on an explicit collection of samples. You could use that approach to maintain your own iteration over multiple batches of a traditional training set but allowing fit or fit_generator to iterate batches for you is likely simpler.
One case when it might be nice to use train_on_batch is for updating a pre-trained model on a single new batch of samples. Suppose you've already trained and deployed a model, and sometime later you've received a new set of training samples previously never used. You could use train_on_batch to directly update the existing model only on those samples. Other methods can do this too, but it is rather explicit to use train_on_batch for this case.
Apart from special cases like this (either where you have some pedagogical reason to maintain your own cursor across different training batches, or else for some type of semi-online training update on a special batch), it is probably better to just always use fit (for data that fits in memory) or fit_generator (for streaming batches of data as a generator).
train_on_batch() gives you greater control of the state of the LSTM, for example, when using a stateful LSTM and controlling calls to model.reset_states() is needed. You may have multi-series data and need to reset the state after each series, which you can do with train_on_batch(), but if you used .fit() then the network would be trained on all the series of data without resetting the state. There's no right or wrong, it depends on what data you're using, and how you want the network to behave.
Train_on_batch will also see a performance increase over fit and fit generator if youre using large datasets and don't have easily serializable data (like high rank numpy arrays), to write to tfrecords.
In this case you can save the arrays as numpy files and load up smaller subsets of them (traina.npy, trainb.npy etc) in memory, when the whole set won't fit in memory. You can then use tf.data.Dataset.from_tensor_slices and then using train_on_batch with your subdataset, then loading up another dataset and calling train on batch again, etc, now you've trained on your entire set and can control exactly how much and what of your dataset trains your model. You can then define your own epochs, batch sizes, etc with simple loops and functions to grab from your dataset.
Indeed #nbro answer helps, just to add few more scenarios, lets say you are training some seq to seq model or a large network with one or more encoders. We can create custom training loops using train_on_batch and use a part of our data to validate on the encoder directly without using callbacks. Writing callbacks for a complex validation process could be difficult. There are several cases where we wish to train on batch.
Regards,
Karthick
From Keras - Model training APIs:
fit: Trains the model for a fixed number of epochs (iterations on a dataset).
train_on_batch: Runs a single gradient update on a single batch of data.
We can use it in GAN when we update the discriminator and generator using a batch of our training data set at a time. I saw Jason Brownlee used train_on_batch in on his tutorials (How to Develop a 1D Generative Adversarial Network From Scratch in Keras)
Tip for quick search: Type Control+F and type in the search box the term that you want to search (train_on_batch, for example).
When developing a neural net one typically partitions training data into Train, Test, and Holdout datasets (many people call these Train, Validation, and Test respectively. Same things, different names). Many people advise selecting hyperparameters based on performance in the Test dataset. My question is: why? Why not maximize performance of hyperparameters in the Train dataset, and stop training the hyperparameters when we detect overfitting via a drop in performance in the Test dataset? Since Train is typically larger than Test, would this not produce better results compared to training hyperparameters on the Test dataset?
UPDATE July 6 2016
Terminology change, to match comment below. Datasets are now termed Train, Validation, and Test in this post. I do not use the Test dataset for training. I am using a GA to optimize hyperparameters. At each iteration of the outer GA training process, the GA chooses a new hyperparameter set, trains on the Train dataset, and evaluates on the Validation and Test datasets. The GA adjusts the hyperparameters to maximize accuracy in the Train dataset. Network training within an iteration stops when network overfitting is detected (in the Validation dataset), and the outer GA training process stops when overfitting of the hyperparameters is detected (again in Validation). The result is hyperparameters psuedo-optimized for the Train dataset. The question is: why do many sources (e.g. https://www.cs.toronto.edu/~hinton/absps/JMLRdropout.pdf, Section B.1) recommend optimizing the hyperparameters on the Validation set, rather than the Train set? Quoting from Srivasta, Hinton, et al (link above): "Hyperparameters were tuned on the validation set such that the best validation error was produced..."
The reason is that developing a model always involves tuning its configuration: for example, choosing the number of layers or the size of the layers (called the hyper-parameters of the model, to distinguish them from the parameters, which are the network’s weights). You do this tuning by using as a feedback signal the performance of the model on the validation data. In essence, this tuning is a form of learning: a search for a good configuration in some parameter space. As a result, tuning the configuration of the model based on its performance on the validation set can quickly result in overfitting to the validation set, even though your model is never directly trained on it.
Central to this phenomenon is the notion of information leaks. Every time you tune a hyperparameter of your model based on the model’s performance on the validation set, some information about the validation data leaks into the model. If you do this only once, for one parameter, then very few bits of information will leak, and your validation set will remain reliable to evaluate the model. But if you repeat this many times—running one experiment, evaluating on the validation set, and modifying your model as a result—then you’ll leak an increasingly significant amount of information about the validation set into the model.
At the end of the day, you’ll end up with a model that performs artificially well on the validation data, because that’s what you optimized it for. You care about performance on completely new data, not the validation data, so you need to use a completely different, never-before-seen dataset to evaluate the model: the test dataset. Your model shouldn’t have had access to any information about the test set, even indirectly. If anything about the model has been tuned based on test set performance, then your measure of generalization will be flawed.
There are two things you are missing here. First, minor, is that test set is never used to do any training. This is a purpose of validation (test is just to asses your final, testing performance). The major missunderstanding is what it means "to use validation set to fit hyperparameters". This means exactly what you describe - to train a model with a given hyperparameters on the training set, and use validation to simply check if you are overfitting (you use it to estimate generalization) , but you do not really "train" on them, you simply check your scores on this subset (which, as you noticed - is way smaller).
You cannot "stop training hyperparamters" because this is not a continuous process, usually hyperparameters are just "possible sets of values", and you have to simply test lots of them, there is no valid way of defining a direct trainingn procedure between actual metric you are interested in (like accuracy) and hyperparameters (like size of the hidden layer in NN or even C parameter in SVM), as the functional link between these two is not differentiable, is highly non convex and in general "ugly" to optimize. If you can define a nice optimization procedure in terms of a hyperparameter than it is usually not called a hyperparameter but a parameter, the crucial distinction in this naming convention is what makes it hard to optimize directly - we call hyperparameter a parameter, than cannot be directly optimized against thus you need a "meta method" (like simply testing on validation set) to select it.
However, you can define a "nice" meta optimization protocol for hyperparameters, but this will still use validation set as an estimator, for example Bayesian optimization of hyperparameters does exactly this - it tries to fit a function saying how well is you model behaving in the space of hyperparameters, but in order to have any "training data" for this meta-method, you need validation set to estimate it for any given set of hyperparameters (input to your meta method)
simple answer: we do
In the case of a simple feedforward neural network you do have to select e.g. layer and unit count per layer, regularization (and non-continuous parameters like topology if not feedforward and loss function) in the beginning and you would optimize on those.
So, in summary you optimize:
ordinary parameters only during training but not during validation
hyperparameters during training and during validation
It is very important not to touch the many ordinary parameters (weights and biases) during validation. That's because there are thousands of degrees of freedom in them which means they can learn the data you train them on. But then the model doesn't generalize to new data as well (even when that new data originated from the same distribution). You usually only have very few degrees of freedom in the hyperparameters which usually control the rigidity of the model (regularization).
This holds true for other machine learning algorithms like decision trees, forests, etc as well.
I've been studying neural networks for a bit and recently learned about the dropout training algorithm. There are excellent papers out there to understand how it works, including the ones from the authors.
So I built a neural network with dropout training (it was fairly easy) but I'm a bit confused about how to perform model selection. From what I understand, looks like dropout is a method to be used when training the final model obtained through model selection.
As for the test part, papers always talk about using the complete network with halved weights, but they do not mention how to use it in the training/validation part (at least the ones I read).
I was thinking about using the network without dropout for the model selection part. Say that makes me find that the net performs well with N neurons. Then, for the final training (the one I use to train the network for the test part) I use 2N neurons with dropout probability p=0.5. That assures me to have exactly N neurons active on average, thus using the network at the right capacity most of the time.
Is this a correct approach?
By the way, I'm aware of the fact that dropout might not be the best choice with small datasets. The project I'm working on has academic purposes, so it's not really needed that I use the best model for the data, as long as I stick with machine learning good practices.
First of all, model selection and the training of a particular model are completely different issues. For model selection, you would usually need a data set that is completely independent of both training set used to build the model and test set used to estimate its performance. So if you're doing for example a cross-validation, you would need an inner cross-validation (to train the models and estimate the performance in general) and an outer cross-validation to do the model selection.
To see why, consider the following thought experiment (shamelessly stolen from this paper). You have a model that makes a completely random prediction. It has a number of parameters that you can set, but have no effect. If you're trying different parameter settings long enough, you'll eventually get a model that has a better performance than all the others simply because you're sampling from a random distribution. If you're using the same data for all of these models, this is the model you will choose. If you have a separate test set, it will quickly tell you that there is no real effect because the performance of this parameter setting that achieves good results during the model-building phase is not better on the separate set.
Now, back to neural networks with dropout. You didn't refer to any particular paper; I'm assuming that you mean Srivastava et. al. "Dropout: A Simple Way to Prevent Neural Networks from Overfitting". I'm not an expert on the subject, but the method to me seems to be similar to what's used in random forests or bagging to mitigate the flaws an individual learner may exhibit by applying it repeatedly in slightly different contexts. If I understood the method correctly, essentially what you end up with is an average over several possible models, very similar to random forests.
This is a way to make an individual model better, but not for model selection. The dropout is a way of adjusting the learned weights for a single neural network model.
To do model selection on this, you would need to train and test neural networks with different parameters and then evaluate those on completely different sets of data, as described in the paper I've referenced above.