Handling features not correlated with output prediction? - machine-learning

I do regression analysis with multiple features. Number of features is 20-23. For now, I check each feature correlation with output variable. Some features show correlation coefficient close to 1 or -1 (highly correlated). Some features show correlation coefficient near 0. My question is: do I have to remove this feature if it has close to 0 correlation coefficient? Or I can keep it and the only problem is that this feature will no make some noticeable effect to regression model or will have faint affect on it. Or removing that kind of features is obligatory?

In short
High (absolute) correlation between a feature and output implies that this feature should be valuable as predictor
Lack of correlation between feature and output implies nothing
More details
Pair-wise correlation only shows you how one thing affects the other, it says completely nothing about how good is this feature connected with others. So if your model is not trivial then you should not drop variables because they are not correlated with output). I will give you the example which should show you why.
Consider following sample, we have 2 features (X, Y), and one output value (Z, say red is 1, black is 0)
X Y Z
1 1 1
1 2 0
1 3 0
2 1 0
2 2 1
2 3 0
3 1 0
3 2 0
3 3 1
Let us compute the correlations:
CORREL(X, Z) = 0
CORREL(Y, Z) = 0
So... we should drop all values? One of them? If we drop any variable - our prolem becomes completely impossible to model! "magic" lies in the fact that there is actually a "hidden" relation in the data.
|X-Y|
0
1
2
1
0
1
2
1
0
And
CORREL(|X-Y|, Z) = -0.8528028654
Now this is a good predictor!
You can actually get a perfect regressor (interpolator) through
Z = 1 - sign(|X-Y|)

Related

Is there a reason why a feature only present in a given class is not being predicted strongly into that class?

Summary & Questions
I'm using liblinear 2.30 - I noticed a similar issue in prod, so I tried to isolate it through a simple reduced training with 2 classes, 1 train doc per class, 5 features with same weight in my vocabulary and 1 simple test doc containing only one feature which is present only in class 2.
a) what's the feature value being used for?
b) I wanted to understand why this test document containing a single feature which is only present in one class is not being strongly predicted into that class?
c) I'm not expecting to have different values per features. Is there any other implications by increasing each feature value from 1 to something-else? How can I determine that number?
d) Could my changes affect other more complex trainings in a bad way?
What I tried
Below you will find data related to a simple training (please focus on feature 5):
> cat train.txt
1 1:1 2:1 3:1
2 2:1 4:1 5:1
> train -s 0 -c 1 -p 0.1 -e 0.01 -B 0 train.txt model.bin
iter 1 act 3.353e-01 pre 3.333e-01 delta 6.715e-01 f 1.386e+00 |g| 1.000e+00 CG 1
iter 2 act 4.825e-05 pre 4.824e-05 delta 6.715e-01 f 1.051e+00 |g| 1.182e-02 CG 1
> cat model.bin
solver_type L2R_LR
nr_class 2
label 1 2
nr_feature 5
bias 0
w
0.3374141436539016
0
0.3374141436539016
-0.3374141436539016
-0.3374141436539016
0
And this is the output of the model:
solver_type L2R_LR
nr_class 2
label 1 2
nr_feature 5
bias 0
w
0.3374141436539016
0
0.3374141436539016
-0.3374141436539016
-0.3374141436539016
0
1 5:10
Below you will find my model's prediction:
> cat test.txt
1 5:1
> predict -b 1 test.txt model.bin test.out
Accuracy = 0% (0/1)
> cat test.out
labels 1 2
2 0.416438 0.583562
And here is where I'm a bit surprised because of the predictions being just [0.42, 0.58] as the feature 5 is only present in class 2. Why?
So I just tried with increasing the feature value for the test doc from 1 to 10:
> cat newtest.txt
1 5:10
> predict -b 1 newtest.txt model.bin newtest.out
Accuracy = 0% (0/1)
> cat newtest.out
labels 1 2
2 0.0331135 0.966887
And now I get a better prediction [0.03, 0.97]. Thus, I tried re-compiling my training again with all features set to 10:
> cat newtrain.txt
1 1:10 2:10 3:10
2 2:10 4:10 5:10
> train -s 0 -c 1 -p 0.1 -e 0.01 -B 0 newtrain.txt newmodel.bin
iter 1 act 1.104e+00 pre 9.804e-01 delta 2.508e-01 f 1.386e+00 |g| 1.000e+01 CG 1
iter 2 act 1.381e-01 pre 1.140e-01 delta 2.508e-01 f 2.826e-01 |g| 2.272e+00 CG 1
iter 3 act 2.627e-02 pre 2.269e-02 delta 2.508e-01 f 1.445e-01 |g| 6.847e-01 CG 1
iter 4 act 2.121e-03 pre 1.994e-03 delta 2.508e-01 f 1.183e-01 |g| 1.553e-01 CG 1
> cat newmodel.bin
solver_type L2R_LR
nr_class 2
label 1 2
nr_feature 5
bias 0
w
0.19420510395364846
0
0.19420510395364846
-0.19420510395364846
-0.19420510395364846
0
> predict -b 1 newtest.txt newmodel.bin newtest.out
Accuracy = 0% (0/1)
> cat newtest.out
labels 1 2
2 0.125423 0.874577
And again predictions were still ok for class 2: 0.87
a) what's the feature value being used for?
Each instance of n features is considered as a point in an n-dimensional space, attached with a given label, say +1 or -1 (in your case 1 or 2). A linear SVM tries to find the best hyperplane to separate those instance into two sets, say SetA and SetB. A hyperplane is considered better than other roughly when SetA contains more instances labeled with +1 and SetB contains more those with -1. i.e., more accurate. The best hyperplane is saved as the model. In your case, the hyperplane has formulation:
f(x)=w^T x
where w is the model, e.g (0.33741,0,0.33741,-0.33741,-0.33741) in your first case.
Probability (for LR) formulation:
prob(x)=1/(1+exp(-y*f(x))
where y=+1 or -1. See Appendix L of LIBLINEAR paper.
b) I wanted to understand why this test document containing a single feature which is only present in one class is not being strongly predicted into that class?
Not only 1 5:1 gives weak probability such as [0.42,0.58], if you predict 2 2:1 4:1 5:1 you will get [0.337417,0.662583] which seems that the solver is also not very confident about the result, even the input is exactly the same as the training data set.
The fundamental reason is the value of f(x), or can be simply seen as the distance between x and the hyperplane. It can be 100% confident x belongs to a certain class only if the distance is infinite large (see prob(x)).
c) I'm not expecting to have different values per features. Is there any other implications by increasing each feature value from 1 to something-else? How can I determine that number?
TL;DR
Enlarging both training and test set is like having a larger penalty parameter C (the -c option). Because larger C means a more strict penalty on error, intuitively speaking, the solver has more confidence with the prediction.
Enlarging every feature of the training set is just like having a smaller C.
Specifically, logistic regression solves the following equation for w.
min 0.5 w^T w + C ∑i log(1+exp(−yi w^T xi))
(eq(3) of LIBLINEAR paper)
For most instance, yi w^T xi is positive and larger xi implies smaller ∑i log(1+exp(−yi w^T xi)).
So the effect is somewhat similar to having a smaller C, and a smaller C implies smaller |w|.
On the other hand, enlarging the test set is the same as having a large |w|. Therefore, the effect of enlarging both training and test set is basically
(1). Having smaller |w| when training
(2). Then, having larger |w| when testing
Because the effect is more dramatic in (2) than (1), overall, enlarging both training and test set is like having a larger |w|, or, having a larger C.
We can run on the data set and multiply every features by 10^12. With C=1, we have the model and probability
> cat model.bin.m1e12.c1
solver_type L2R_LR
nr_class 2
label 1 2
nr_feature 5
bias 0
w
3.0998430106024949e-12
0
3.0998430106024949e-12
-3.0998430106024949e-12
-3.0998430106024949e-12
0
> cat test.out.m1e12.c1
labels 1 2
2 0.0431137 0.956886
Next we run on the original data set. With C=10^12, we have the probability
> cat model.bin.m1.c1e12
solver_type L2R_LR
nr_class 2
label 1 2
nr_feature 5
bias 0
w
3.0998430101989314
0
3.0998430101989314
-3.0998430101989314
-3.0998430101989314
0
> cat test.out.m1.c1e12
labels 1 2
2 0.0431137 0.956886
Therefore, because larger C means more strict penalty on error, so intuitively the solver has more confident with prediction.
d) Could my changes affect other more complex trainings in a bad way?
From (c) we know your changes is like having a larger C, and that will result in a better training accuracy. But it almost can be sure that the model is over fitting the training set when C goes too large. As a result, the model cannot endure the noise in training set and will perform badly in test accuracy.
As for finding a good C, a popular way is by cross validation (-v option).
Finally,
it may be off-topic but you may want to see how to pre-process the text data. It is common (e.g., suggested by the author of liblinear here) to instance-wise normalize the data.
For document classification, our experience indicates that if you normalize each document to unit length, then not only the training time is shorter, but also the performance is better.

How to remove the outliers using Python

I am doing a binary classification problem, I am struggling with removing outliers and also increasing accuracy.
Ratings are one my feature looks like this:
0 0.027465
1 0.027465
2 0.027465
3 0.027465
4 0.027465
...
26043 0.027465
26044 0.027465
26045 0.102234
26046 0.027465
26047 0.027465
mean value of the data:
train.ratings.mean()
0.03871552285960927
std of the data:
train.ratings.std()
0.07585168664836195
I tried the log transformation but accuracy is not increased:
train['ratings']=np.log(train.ratings+1)
my goal is to classify the data true or false:
train.netgain
0 False
1 False
2 False
3 False
4 True
...
26043 True
26044 False
26045 True
26046 False
26047 Fals
One method I used was to calculate a MAD and after that I tag all outlier with a bool type with that I can get all outliers.
Sample of MAD calculation:
def mad(x):
return np.median(np.abs(x - np.median(x)))
def mad_ratio(x):
mad_value = mad(x)
if mad_value == 0:
return 0
x_mad = np.abs(x - np.median(x)) / mad_value
return x_mad
Assume that the rating feature is normally distributed and convert it to the standard normal distribution
From normal distribution, we know 99.7% values are covered with 3 standard deviations. so we can remove the values which are above 3 standard deviations away from the mean.
.**
See below for python code.
ratings_mean=train['ratings'].mean() #Finding the mean of ratings column
ratings_std=train['ratings'].std() # standard deviation of the column
train['ratings']=train['ratings'].map(lamdba x: (x - ratings_mean)/ ratings_std
Ok, now we have now converted our data into a standard normal distribution. Now we if you see, its mean should be 0 and the standard deviation should be 1. From this, we can find out which are greater than 3 and less than -3. so that we can remove those rows from the dataset.
train=train[np.abs(train_ratings) < 3]
Now train dataframe will remove the outliers from the dataset.
**Note: You can apply 2 standard deviations as well because 2-std contains 95% of the data. Its all depends on the domain knowledge and your data. **

Encode a categorical feature with multiple categories per example

I am working on a dataset which has a feature that has multiple categories for a single example.
The feature looks like this:-
Feature
0 [Category1, Category2, Category2, Category4, Category5]
1 [Category11, Category20, Category133]
2 [Category2, Category9]
3 [Category1000, Category1200, Category2000]
4 [Category12]
The problem is similar to the this question posted:- Encode categorical features with multiple categories per example - sklearn
Now, I want to vectorize this feature. One solution is to use MultiLabelBinarizer as suggested in the answer of the above similar question. But, there are around 2000 categories, which results into a sparse and very high dimentional encoded data.
Is there any other encoding that can be used? Or any possible solution for this problem. Thanks.
Given an incredibly sparse array one could use a dimensionality reduction technique such as PCA (Principal component analysis) to reduce the feature space to the top k features that best describe the variance.
Assuming the MultiLabelBinarizered 2000 features = X
from sklearn.decomposition import PCA
k = 5
model = PCA(n_components = k, random_state = 666)
model.fit(X)
Components = model.predict(X)
And then you can use the top K components as a smaller dimensional feature space that can explain a large portion of the variance for the original feature space.
If you want to understand how well the new smaller feature space describes the variance you could use the following command
model.explained_variance_
In many cases when I encountered the problem of too many features being generated from a column with many categories, I opted for binary encoding and it worked out fine most of the times and hence is worth a shot for you perhaps.
Imagine you have 9 features, and you mark them from 1 to 9 and now binary encode them, you will get:
cat 1 - 0 0 0 1
cat 2 - 0 0 1 0
cat 3 - 0 0 1 1
cat 4 - 0 1 0 0
cat 5 - 0 1 0 1
cat 6 - 0 1 1 0
cat 7 - 0 1 1 1
cat 8 - 1 0 0 0
cat 9 - 1 0 0 1
This is the basic intuition behind Binary Encoder.
PS: Given that 2 power 11 is 2048 and you may have 2000 categories or so, you can reduce your categories to 11 feature columns instead of many (for example, 1999 in the case of one-hot)!
I also encountered these same problems but I solved using Countvectorizer from sklearn.feature_extraction.text just by giving binary=True, i.e CounterVectorizer(binary=True)

Data Science: Scoring methodology

I am looking for any methodology to assign a risk score to an individual based on certain events. I am looking to have a 0-100 scale with an exponential assignment. For example, for one event a day the score may rise to 25, for 2 it may rise to 50-60 and for 3-4 events a day the score for the day would be 100.
I tried to Google it but since I am not aware of the right terminology, I am landing up on random topics. :(
Is there any mathematical terminology for this kind of scoring system? what are the most common methods you might know?
P.S.: Expert/experience data scientist advice highly appreciated ;)
I would start by writing some qualifications:
0 events trigger a score of 0.
Non edge event count observations are where the score – 100-threshold would live.
Any score after the threshold will be 100.
If so, here's a (very) simplified example:
Stage Data:
userid <- c("a1","a2","a3","a4","a11","a12","a13","a14","u2","wtf42","ub40","foo","bar","baz","blue","bop","bob","boop","beep","mee","r")
events <- c(0,0,0,0,0,0,0,0,0,0,0,0,1,2,3,2,3,6,122,13,1)
df1 <- data.frame(userid,events)
Optional: Normalize events to be in (1,2].
This might be helpful for logarithmic properties. (Otherwise, given the assumed function, score=events^exp, as in this example, 1 event will always yield a score of 1) This will allow you to control sensitivity, but it must be done right as we are dealing with exponents and logarithms. I am not using normalization in the example:
normevents <- (events-mean(events))/((max(events)-min(events))*2)+1.5
Set the quantile threshold for max score:
MaxScoreThreshold <- 0.25
Get the non edge quintiles of the events distribution:
qts <- quantile(events[events>min(events) & events<max(events)], c(seq(from=0, to=100,by=5)/100))
Find the Events quantity that give a score of 100 using the set threshold.
MaxScoreEvents <- quantile(qts,MaxScoreThreshold)
Find the exponent of your exponential function
Given that:
Score = events ^ exponent
events is a Natural number - integer >0: We took care of it by
omitting the edges)
exponent > 1
Exponent Calculation:
exponent <- log(100)/log(MaxScoreEvents)
Generate the scores:
df1$Score <- apply(as.matrix(events^exponent),1,FUN = function(x) {
if (x > 100) {
result <- 100
}
else if (x < 0) {
result <- 0
}
else {
result <- x
}
return(ceiling(result))
})
df1
Resulting Data Frame:
userid events Score
1 a1 0 0
2 a2 0 0
3 a3 0 0
4 a4 0 0
5 a11 0 0
6 a12 0 0
7 a13 0 0
8 a14 0 0
9 u2 0 0
10 wtf42 0 0
11 ub40 0 0
12 foo 0 0
13 bar 1 1
14 baz 2 100
15 blue 3 100
16 bop 2 100
17 bob 3 100
18 boop 6 100
19 beep 122 100
20 mee 13 100
21 r 1 1
Under the assumption that your data is larger and has more event categories, the score won't snap to 100 so quickly, it is also a function of the threshold.
I would rely more on the data to define the parameters, threshold in this case.
If you have prior data as to what users really did whatever it is your score assess you can perform supervised learning, set the threshold # wherever the ratio is over 50% for example. Or If the graph of events to probability of ‘success’ looks like the cumulative probability function of a normal distribution, I’d set threshold # wherever it hits 45 degrees (For the first time).
You could also use logistic regression if you have prior data but instead of a Logit function ingesting the output of regression, use the number as your score. You can normalize it to be within 0-100.
It’s not always easy to write a Data Science question. I made many assumptions as to what you are looking for, hope this is the general direction.

Padding time-series subsequences for LSTM-RNN training

I have a dataset of time series that I use as input to an LSTM-RNN for action anticipation. The time series comprises a time of 5 seconds at 30 fps (i.e. 150 data points), and the data represents the position/movement of facial features.
I sample additional sub-sequences of smaller length from my dataset in order to add redundancy in the dataset and reduce overfitting. In this case I know the starting and ending frame of the sub-sequences.
In order to train the model in batches, all time series need to have the same length, and according to many papers in the literature padding should not affect the performance of the network.
Example:
Original sequence:
1 2 3 4 5 6 7 8 9 10
Subsequences:
4 5 6 7
8 9 10
2 3 4 5 6
considering that my network is trying to anticipate an action (meaning that as soon as P(action) > threshold as it goes from t = 0 to T = tmax, it will predict that action) will it matter where the padding goes?
Option 1: Zeros go to substitute original values
0 0 0 4 5 6 7 0 0 0
0 0 0 0 0 0 0 8 9 10
0 2 3 4 5 6 0 0 0 0
Option 2: all zeros at the end
4 5 6 7 0 0 0 0 0 0
8 9 10 0 0 0 0 0 0 0
2 3 4 5 0 0 0 0 0 0
Moreover, some of the time series are missing a number of frames, but it is not known which ones they are - meaning that if we only have 60 frames, we don't know whether they are taken from 0 to 2 seconds, from 1 to 3s, etc. These need to be padded before the subsequences are even taken. What is the best practice for padding in this case?
Thank you in advance.
The most powerful attribute of LSTMs and RNNs in general is that their parameters are shared along the time frames(Parameters recur over time frames) but the parameter sharing relies upon the assumption that the same parameters can be used for different time steps i.e. the relationship between the previous time step and the next time step does not depend on t as explained here in page 388, 2nd paragraph.
In short, padding zeros at the end, theoretically should not change the accuracy of the model. I used the adverb theoretically because at each time step LSTM's decision depends on its cell state among other factors and this cell state is kind of a short summary of the past frames. As far as I understood, that past frames may be missing in your case. I think what you have here is a little trade-off.
I would rather pad zeros at the end because it doesn't completely conflict with the underlying assumption of RNNs and it's more convenient to implement and keep track of.
On the implementation side, I know tensorflow calculates the loss function once you give it the sequences and the actual sequence size of each sample(e.g. for 4 5 6 7 0 0 0 0 0 0 you also need to give it the actual size which is 4 here) assuming you're implementing the option 2. I don't know whether there is an implementation for option 1, though.
Better go for padding zeroes in the beginning, as this paper suggests Effects of padding on LSTMs and CNNs,
Though post padding model peaked it’s efficiency at 6 epochs and started to overfit after that, it’s accuracy is way less than pre-padding.
Check table 1, where the accuracy of pre-padding(padding zeroes in the beginning) is around 80%, but for post-padding(padding zeroes in the end), it is only around 50%
In case you have sequences of variable length, pytorch provides a utility function torch.nn.utils.rnn.pack_padded_sequence. The general workflow with this function is
from torch.nn.utils.rnn import pack_padded_sequence, pad_packed_sequence
embedding = nn.Embedding(4, 5)
rnn = nn.GRU(5, 5)
sequences = torch.tensor([[1,2,0], [3,0,0], [2,1,3]])
lens = [2, 1, 3] # indicating the actual length of each sequence
embeddings = embedding(sequences)
packed_seq = pack_padded_sequence(embeddings, lens, batch_first=True, enforce_sorted=False)
e, hn = rnn(packed_seq)
One can collect the embedding of each token by
e = pad_packed_sequence(e, batch_first=True)
Using this function is better than padding by yourself, because torch will limit RNN to only inspecting the actual sequence and stop before the padded token.

Resources