I have a classification task. The training data has 50 different labels. The customer wants to differentiate the low probability predictions, meaning that, I have to classify some test data as Unclassified / Other depending on the probability (certainty?) of the model.
When I test my code, the prediction result is a numpy array (I'm using different models, this is one is pre-trained BertTransformer). The prediction array doesn't contain probabilities such as in Keras predict_proba() method. These are numbers generated by prediction method of pretrained BertTransformer model.
[[-1.7862008 -0.7037363 0.09885322 1.5318055 2.1137428 -0.2216074
0.18905772 -0.32575375 1.0748093 -0.06001111 0.01083148 0.47495762
0.27160102 0.13852511 -0.68440574 0.6773654 -2.2712054 -0.2864312
-0.8428862 -2.1132915 -1.0157436 -1.0340284 -0.35126117 -1.0333195
9.149789 -0.21288703 0.11455813 -0.32903734 0.10503325 -0.3004114
-1.3854568 -0.01692022 -0.4388664 -0.42163098 -0.09182278 -0.28269592
-0.33082992 -1.147654 -0.6703184 0.33038092 -0.50087476 1.1643585
0.96983343 1.3400391 1.0692116 -0.7623776 -0.6083422 -0.91371405
0.10002492]]
I'm using numpy.argmax() to identify the correct label. The prediction works just fine. However, since these are not probabilities, I cannot compare the best result with a threshold value.
My question is, how can I define a threshold (say, 0.6), and then compare the probability of the argmax() element of the BertTransformer prediction array so that I can classify the prediction as "Other" if the probability is less than the threshold value?
Edit 1:
We are using 2 different models. One is Keras, and the other is BertTransformer. We have no problem in Keras since it gives the probabilities so I'm skipping Keras model.
The Bert model is pretrained. Here is how it is generated:
def model(self, data):
number_of_categories = len(data['encoded_categories'].unique())
model = BertForSequenceClassification.from_pretrained(
"dbmdz/bert-base-turkish-128k-uncased",
num_labels=number_of_categories,
output_attentions=False,
output_hidden_states=False,
)
# model.cuda()
return model
The output given above is the result of model.predict() method. We compare both models, Bert is slightly ahead, therefore we know that the prediction works just fine. However, we are not sure what those numbers signify or represent.
Here is the Bert documentation.
BertForSequenceClassification returns logits, i.e., the classification scores before normalization. You can normalize the scores by calling F.softmax(output, dim=-1) where torch.nn.functional was imported as F.
With thousands of labels, the normalization can be costly and you do not need it when you are only interested in argmax. This is probably why the models return the raw scores only.
I have trained LightGBM on a binary-classification problem, and when plotting the tree I get some leafs like this
I struggle to find the loss-function for the classification trees - Does LightGBM minimize the cross-entropy in the binary case, and is that the leaf score?
I struggle to find the loss-function for the classification trees - Does LightGBM minimize the cross-entropy in the binary case
Yes, if you don't specify an objective then LGBMClassifier will use cross-entropy by default. The documentation in https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.LGBMClassifier.html#lightgbm.LGBMClassifier says that the default for objective is "binary", and then https://lightgbm.readthedocs.io/en/latest/Parameters.html#objective notes that binary is cross-entropy loss.
and is that the leaf score?
The values like leaf 33: -2.209 ("leaf scores") represent the value of the target that will be predicted for instances in that leaf node, multiplied by the learning rate.
Negative values are possible because of the way the boosting process works. Each tree is trained on the residuals of the model up to that tree. A prediction from a model is obtained by summing the output of all trees. The XGBoost docs have a very good explanation of this: "Introduction to Boosted Trees".
In the future, please try to provide a small reproducible example explaining how you created a figure that you're asking questions about. I assumed something like the following Python code, using lightgbm 3.1.0. You can change the values of tree_index to see the different trees in the model.
import lightgbm as lgb
from sklearn.datasets import load_breast_cancer
X, y = load_breast_cancer(return_X_y=True)
gbm = lgb.LGBMClassifier(
n_estimators=10,
num_leaves=3,
max_depth=8,
min_data_in_leaf=3,
)
gbm.fit(X, y)
# visualize tree structure as a directed graph
ax = lgb.plot_tree(
gbm,
tree_index=0,
figsize=(15, 8),
show_info=[
'data_percentage',
]
)
# visualize tree structure in a dataframe
gbm.booster_.trees_to_dataframe()
I have a classification problem with 10 features and I have to predict 1 or 0. When I train the SVC model, with the train test split, all the predicted values for the test portion of the data comes out to be 0. The data has the following 0-1 count:
0: 1875
1: 1463
The code to train the model is given below:
from sklearn.svm import SVC
model = SVC()
model.fit(X_train, y_train)
pred= model.predict(X_test)
from sklearn.metrics import accuracy_score
accuracy_score(y_test, pred)`
Why does it predict 0 for all the cases?
The model predicts the more frequent class, even though the dataset is nor much imbalanced. It is very likely that the class cannot be predicted from the features as they are right now.
You may try normalizing the features.
Another thing you might want to try is to have a look at how correlated the features are with each other. Having highly correlated features might also prevent the model from converging.
Also, you might have chosen the wrong features.
For a classification problem, it is always good to run a dummy classifiar as a starting point. This will give you an idea how good your model can be.
You can use this as a code:
from sklearn.dummy import DummyClassifier
dummy_classifier = DummyClassifier(strategy="most_frequent")
dummy_classifier.fit(X_train,y_train)
pred_dum= dummy_classifier.predict(X_test)
accuracy_score(y_test, pred_dum)
this will give you an accuracy, if you predict always the most frequent class. If this is for example: 100% , this would mean that you only have one class in your dataset. 80% means, that 80% of your data belongs to one class.
In a first step you can adjust your SVC:
model = SVC(C=1.0, kernel=’rbf’, random_state=42)
C : float, optional (default=1.0)Penalty parameter C of the error
term.
kernel : Specifies the kernel type to be used in the algorithm. It
must be one of ‘linear’, ‘poly’, ‘rbf’
This can give you a starting point.
On top you should run also a prediction for your training data, to see the comparison if you are over- or underfitting.
trainpred= model.predict(X_train)
accuracy_score(y_test, trainpred)
So far I have resourced another post and sklearn documentation
So in general I want to produce the following example:
X = np.matrix([[1,2],[2,3],[3,4],[4,5]])
y = np.array(['A', 'B', 'B', 'C', 'D'])
Xt = np.matrix([[11,22],[22,33],[33,44],[44,55]])
model = model.fit(X, y)
pred = model.predict(Xt)
However for output, I would like to see 3 columns per observation as output from pred:
A | B | C
.5 | .2 | .3
.25 | .25 | .5
...
and a different probability for each class showing up in my prediction.
I believe that the best approach would be Multilabel classification from the second link I provided above. Additionally, I think it might be a good idea to hop into one of the multi-label or multi-output models listed below:
Support multilabel:
sklearn.tree.DecisionTreeClassifier
sklearn.tree.ExtraTreeClassifier
sklearn.ensemble.ExtraTreesClassifier
sklearn.neighbors.KNeighborsClassifier
sklearn.neural_network.MLPClassifier
sklearn.neighbors.RadiusNeighborsClassifier
sklearn.ensemble.RandomForestClassifier
sklearn.linear_model.RidgeClassifierCV
Support multiclass-multioutput:
sklearn.tree.DecisionTreeClassifier
sklearn.tree.ExtraTreeClassifier
sklearn.ensemble.ExtraTreesClassifier
sklearn.neighbors.KNeighborsClassifier
sklearn.neighbors.RadiusNeighborsClassifier
sklearn.ensemble.RandomForestClassifier
However, I am looking for someone who is has more confidence and experience at doing this the right way. All feedback is appreciated.
-bmc
From what I understand you want to obtain probabilities for each of the potential classes for multi-class classifier.
In Scikit-Learn it can be done by generic function predict_proba. It is implemented for most of the classifiers in scikit-learn. You basically call:
clf.predict_proba(X)
Where clf is the trained classifier.
As output you will get a decimal array of probabilities for each class for each input value.
One word of caution - not all classifiers naturally evaluate class probabilities. For instance, SVM doesn't do that. You still can obtain the class probabilities though, but to do that upon constructing such classifiers you need to instruct it to perform probability estimation. For SVM it would look like:
SVC(Probability=True)
After you fit it you will be able to use predict_proba as before.
I need to warn you that if classifier doesn't naturally evaluate probabilities that means that the probabilities will be evaluated using rather expansive computational methods which may significantly increase training time. So I advice you to use classifiers which naturally evaluate class probabilities (neural networks with softmax output, logistic regression, gradient boosting etc)
Try to use calibrated model:
# define model
model = SVC()
# define and fit calibration model
calibrated = CalibratedClassifierCV(model, method='sigmoid', cv=5)
calibrated.fit(trainX, trainy)
# predict probabilities
print(calibrated.predict_proba(testX)[:, 1])
I'm trying to train a CNN to categorize text by topic. When I use binary cross-entropy I get ~80% accuracy, with categorical cross-entropy I get ~50% accuracy.
I don't understand why this is. It's a multiclass problem, doesn't that mean that I have to use categorical cross-entropy and that the results with binary cross-entropy are meaningless?
model.add(embedding_layer)
model.add(Dropout(0.25))
# convolution layers
model.add(Conv1D(nb_filter=32,
filter_length=4,
border_mode='valid',
activation='relu'))
model.add(MaxPooling1D(pool_length=2))
# dense layers
model.add(Flatten())
model.add(Dense(256))
model.add(Dropout(0.25))
model.add(Activation('relu'))
# output layer
model.add(Dense(len(class_id_index)))
model.add(Activation('softmax'))
Then I compile it either it like this using categorical_crossentropy as the loss function:
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
or
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
Intuitively it makes sense why I'd want to use categorical cross-entropy, I don't understand why I get good results with binary, and poor results with categorical.
The reason for this apparent performance discrepancy between categorical & binary cross entropy is what user xtof54 has already reported in his answer below, i.e.:
the accuracy computed with the Keras method evaluate is just plain
wrong when using binary_crossentropy with more than 2 labels
I would like to elaborate more on this, demonstrate the actual underlying issue, explain it, and offer a remedy.
This behavior is not a bug; the underlying reason is a rather subtle & undocumented issue at how Keras actually guesses which accuracy to use, depending on the loss function you have selected, when you include simply metrics=['accuracy'] in your model compilation. In other words, while your first compilation option
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
is valid, your second one:
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
will not produce what you expect, but the reason is not the use of binary cross entropy (which, at least in principle, is an absolutely valid loss function).
Why is that? If you check the metrics source code, Keras does not define a single accuracy metric, but several different ones, among them binary_accuracy and categorical_accuracy. What happens under the hood is that, since you have selected binary cross entropy as your loss function and have not specified a particular accuracy metric, Keras (wrongly...) infers that you are interested in the binary_accuracy, and this is what it returns - while in fact you are interested in the categorical_accuracy.
Let's verify that this is the case, using the MNIST CNN example in Keras, with the following modification:
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) # WRONG way
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=2, # only 2 epochs, for demonstration purposes
verbose=1,
validation_data=(x_test, y_test))
# Keras reported accuracy:
score = model.evaluate(x_test, y_test, verbose=0)
score[1]
# 0.9975801164627075
# Actual accuracy calculated manually:
import numpy as np
y_pred = model.predict(x_test)
acc = sum([np.argmax(y_test[i])==np.argmax(y_pred[i]) for i in range(10000)])/10000
acc
# 0.98780000000000001
score[1]==acc
# False
To remedy this, i.e. to use indeed binary cross entropy as your loss function (as I said, nothing wrong with this, at least in principle) while still getting the categorical accuracy required by the problem at hand, you should ask explicitly for categorical_accuracy in the model compilation as follows:
from keras.metrics import categorical_accuracy
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=[categorical_accuracy])
In the MNIST example, after training, scoring, and predicting the test set as I show above, the two metrics now are the same, as they should be:
# Keras reported accuracy:
score = model.evaluate(x_test, y_test, verbose=0)
score[1]
# 0.98580000000000001
# Actual accuracy calculated manually:
y_pred = model.predict(x_test)
acc = sum([np.argmax(y_test[i])==np.argmax(y_pred[i]) for i in range(10000)])/10000
acc
# 0.98580000000000001
score[1]==acc
# True
System setup:
Python version 3.5.3
Tensorflow version 1.2.1
Keras version 2.0.4
UPDATE: After my post, I discovered that this issue had already been identified in this answer.
It all depends on the type of classification problem you are dealing with. There are three main categories
binary classification (two target classes),
multi-class classification (more than two exclusive targets),
multi-label classification (more than two non exclusive targets), in which multiple target classes can be on at the same time.
In the first case, binary cross-entropy should be used and targets should be encoded as one-hot vectors.
In the second case, categorical cross-entropy should be used and targets should be encoded as one-hot vectors.
In the last case, binary cross-entropy should be used and targets should be encoded as one-hot vectors. Each output neuron (or unit) is considered as a separate random binary variable, and the loss for the entire vector of outputs is the product of the loss of single binary variables. Therefore it is the product of binary cross-entropy for each single output unit.
The binary cross-entropy is defined as
and categorical cross-entropy is defined as
where c is the index running over the number of classes C.
I came across an "inverted" issue — I was getting good results with categorical_crossentropy (with 2 classes) and poor with binary_crossentropy. It seems that problem was with wrong activation function. The correct settings were:
for binary_crossentropy: sigmoid activation, scalar target
for categorical_crossentropy: softmax activation, one-hot encoded target
It's really interesting case. Actually in your setup the following statement is true:
binary_crossentropy = len(class_id_index) * categorical_crossentropy
This means that up to a constant multiplication factor your losses are equivalent. The weird behaviour that you are observing during a training phase might be an example of a following phenomenon:
At the beginning the most frequent class is dominating the loss - so network is learning to predict mostly this class for every example.
After it learnt the most frequent pattern it starts discriminating among less frequent classes. But when you are using adam - the learning rate has a much smaller value than it had at the beginning of training (it's because of the nature of this optimizer). It makes training slower and prevents your network from e.g. leaving a poor local minimum less possible.
That's why this constant factor might help in case of binary_crossentropy. After many epochs - the learning rate value is greater than in categorical_crossentropy case. I usually restart training (and learning phase) a few times when I notice such behaviour or/and adjusting a class weights using the following pattern:
class_weight = 1 / class_frequency
This makes loss from a less frequent classes balancing the influence of a dominant class loss at the beginning of a training and in a further part of an optimization process.
EDIT:
Actually - I checked that even though in case of maths:
binary_crossentropy = len(class_id_index) * categorical_crossentropy
should hold - in case of keras it's not true, because keras is automatically normalizing all outputs to sum up to 1. This is the actual reason behind this weird behaviour as in case of multiclassification such normalization harms a training.
After commenting #Marcin answer, I have more carefully checked one of my students code where I found the same weird behavior, even after only 2 epochs ! (So #Marcin's explanation was not very likely in my case).
And I found that the answer is actually very simple: the accuracy computed with the Keras method evaluate is just plain wrong when using binary_crossentropy with more than 2 labels. You can check that by recomputing the accuracy yourself (first call the Keras method "predict" and then compute the number of correct answers returned by predict): you get the true accuracy, which is much lower than the Keras "evaluate" one.
a simple example under a multi-class setting to illustrate
suppose you have 4 classes (onehot encoded) and below is just one prediction
true_label = [0,1,0,0]
predicted_label = [0,0,1,0]
when using categorical_crossentropy, the accuracy is just 0 , it only cares about if you get the concerned class right.
however when using binary_crossentropy, the accuracy is calculated for all classes, it would be 50% for this prediction. and the final result will be the mean of the individual accuracies for both cases.
it is recommended to use categorical_crossentropy for multi-class(classes are mutually exclusive) problem but binary_crossentropy for multi-label problem.
As it is a multi-class problem, you have to use the categorical_crossentropy, the binary cross entropy will produce bogus results, most likely will only evaluate the first two classes only.
50% for a multi-class problem can be quite good, depending on the number of classes. If you have n classes, then 100/n is the minimum performance you can get by outputting a random class.
You are passing a target array of shape (x-dim, y-dim) while using as loss categorical_crossentropy. categorical_crossentropy expects targets to be binary matrices (1s and 0s) of shape (samples, classes). If your targets are integer classes, you can convert them to the expected format via:
from keras.utils import to_categorical
y_binary = to_categorical(y_int)
Alternatively, you can use the loss function sparse_categorical_crossentropy instead, which does expect integer targets.
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
when using the categorical_crossentropy loss, your targets should be in categorical format (e.g. if you have 10 classes, the target for each sample should be a 10-dimensional vector that is all-zeros except for a 1 at the index corresponding to the class of the sample).
Take a look at the equation you can find that binary cross entropy not only punish those label = 1, predicted =0, but also label = 0, predicted = 1.
However categorical cross entropy only punish those label = 1 but predicted = 1.That's why we make assumption that there is only ONE label positive.
The main point is answered satisfactorily with the brilliant piece of sleuthing by desernaut. However there are occasions when BCE (binary cross entropy) could throw different results than CCE (categorical cross entropy) and may be the preferred choice. While the thumb rules shared above (which loss to select) work fine for 99% of the cases, I would like to add a few new dimensions to this discussion.
The OP had a softmax activation and this throws a probability distribution as the predicted value. It is a multi-class problem. The preferred loss is categorical CE. Essentially this boils down to -ln(p) where 'p' is the predicted probability of the lone positive class in the sample. This means that the negative predictions dont have a role to play in calculating CE. This is by intention.
On a rare occasion, it may be needed to make the -ve voices count. This can be done by treating the above sample as a series of binary predictions. So if expected is [1 0 0 0 0] and predicted is [0.1 0.5 0.1 0.1 0.2], this is further broken down into:
expected = [1,0], [0,1], [0,1], [0,1], [0,1]
predicted = [0.1, 0.9], [.5, .5], [.1, .9], [.1, .9], [.2, .8]
Now we proceed to compute 5 different cross entropies - one for each of the above 5 expected/predicted combo and sum them up. Then:
CE = -[ ln(.1) + ln(0.5) + ln(0.9) + ln(0.9) + ln(0.8)]
The CE has a different scale but continues to be a measure of the difference between the expected and predicted values. The only difference is that in this scheme, the -ve values are also penalized/rewarded along with the +ve values. In case your problem is such that you are going to use the output probabilities (both +ve and -ves) instead of using the max() to predict just the 1 +ve label, then you may want to consider this version of CE.
How about a multi-label situation where expected = [1 0 0 0 1]? Conventional approach is to use one sigmoid per output neuron instead of an overall softmax. This ensures that the output probabilities are independent of each other. So we get something like:
expected = [1 0 0 0 1]
predicted is = [0.1 0.5 0.1 0.1 0.9]
By definition, CE measures the difference between 2 probability distributions. But the above two lists are not probability distributions. Probability distributions should always add up to 1. So conventional solution is to use same loss approach as before - break the expected and predicted values into 5 individual probability distributions, proceed to calculate 5 cross entropies and sum them up. Then:
CE = -[ ln(.1) + ln(0.5) + ln(0.9) + ln(0.9) + ln(0.9)] = 3.3
The challenge happens when the number of classes may be very high - say a 1000 and there may be only couple of them present in each sample. So the expected is something like: [1,0,0,0,0,0,1,0,0,0.....990 zeroes]. The predicted could be something like: [.8, .1, .1, .1, .1, .1, .8, .1, .1, .1.....990 0.1's]
In this case the CE =
- [ ln(.8) + ln(.8) for the 2 +ve classes and 998 * ln(0.9) for the 998 -ve classes]
= 0.44 (for the +ve classes) + 105 (for the negative classes)
You can see how the -ve classes are beginning to create a nuisance value when calculating the loss. The voice of the +ve samples (which may be all that we care about) is getting drowned out. What do we do? We can't use categorical CE (the version where only +ve samples are considered in calculation). This is because, we are forced to break up the probability distributions into multiple binary probability distributions because otherwise it would not be a probability distribution in the first place. Once we break it into multiple binary probability distributions, we have no choice but to use binary CE and this of course gives weightage to -ve classes.
One option is to drown the voice of the -ve classes by a multiplier. So we multiply all -ve losses by a value gamma where gamma < 1. Say in above case, gamma can be .0001. Now the loss comes to:
= 0.44 (for the +ve classes) + 0.105 (for the negative classes)
The nuisance value has come down. 2 years back Facebook did that and much more in a paper they came up with where they also multiplied the -ve losses by p to the power of x. 'p' is the probability of the output being a +ve and x is a constant>1. This penalized -ve losses even further especially the ones where the model is pretty confident (where 1-p is close to 1). This combined effect of punishing negative class losses combined with harsher punishment for the easily classified cases (which accounted for majority of the -ve cases) worked beautifully for Facebook and they called it focal loss.
So in response to OP's question of whether binary CE makes any sense at all in his case, the answer is - it depends. In 99% of the cases the conventional thumb rules work but there could be occasions when these rules could be bent or even broken to suit the problem at hand.
For a more in-depth treatment, you can refer to: https://towardsdatascience.com/cross-entropy-classification-losses-no-math-few-stories-lots-of-intuition-d56f8c7f06b0
The binary_crossentropy(y_target, y_predict) doesn't need to apply to binary classification problem.
In the source code of binary_crossentropy(), the nn.sigmoid_cross_entropy_with_logits(labels=target, logits=output) of tensorflow was actually used.
And, in the documentation, it says that:
Measures the probability error in discrete classification tasks in which each class is independent and not mutually exclusive. For instance, one could perform multilabel classification where a picture can contain both an elephant and a dog at the same time.