What is Active Learning?
Active Learning is a special case of Supervised Machine Learning. This approach is used to construct a high-performance classifier while keeping the size of the training dataset to a minimum by actively selecting the valuable data points.
Where should we apply active learning?
- We have a very small amount or a huge amount of dataset.
- Annotation of the unlabeled dataset cost human effort, time, and money.
- We have access to limited processing power.
Example
On a certain planet, there are various fruits of different size(1-5), some of them are poisonous and others don’t. The only criteria to decide a fruit is poisonous or not is its size. our task is to train a classifier that predicts the given fruit is poisonous or not. The only information we have, is a fruit with size 1 is not poisonous, the fruit of size 5 is poisonous and after a particular size, all fruits are poisonous.
The first approach is to check each and every size of the fruit, which consumes time and resources.
The second approach is to apply the binary search and find the transition point (decision boundary). This approach uses fewer data and gives the same results as of linear search.
General Algorithm : 1. train classifier with the initial training dataset 2. calculate the accuracy 3. while(accuracy < desired accuracy): 4. select the most valuable data points (in general points close to decision boundary) 5. query that data point/s (ask for a label) from human oracle 6. add that data point/s to our initial training dataset 7. re-train the model 8. re-calculate the accuracy
Approaches Active Learning Algorithm
1. Query Synthesis
- Generally, this approach is used when we have a very small dataset.
- This approach we choose any uncertain point from given n-dimensional space. we don’t care about the existence of that point.
In this query, synthesis can pick any point(valuable) from 3*3 2-D plane.
- Sometimes it would be difficult for human oracle to annotate the queried data point.
These are some queries generated by the Query Synthesis approach for a model trained for handwritten recognition. It is very difficult to annotate these queries.
2. Sampling
- This approach is used when we have a large dataset.
- In this approach, we split our dataset into three parts: Training Set; Test Set; Unlabeled Pool(ironical) [5%; 25%, 70%].
- This training dataset is our initial dataset and is used to initially train our model.
- This approach selects valuable/uncertain points from this unlabeled pool, this ensures that all the query can be recognized by human oracle
Black points represents unlabeled pool and union red, green colour dots represents training dataset.
Here is an active learning model which decides valuable points on the basis of, the probability of a point present in a class. In Logistic Regression points closest to the threshold (i.e. probability = 0.5) is the most uncertain point. So, I choose the probability between 0.47 to 0.53 as a range of uncertainty.
You can download the dataset from here.
Python3
import numpy as npimport pandas as pdfrom statistics import meanfrom sklearn.impute import SimpleImputerfrom sklearn.preprocessing import StandardScalerfrom sklearn.linear_model import LogisticRegressionfrom sklearn.model_selection import train_test_split# split dataset into test set, train set and unlabel pooldef split(dataset, train_size, test_size): x = dataset[:, :-1] y = dataset[:, -1] x_train, x_pool, y_train, y_pool = train_test_split( x, y, train_size = train_size) unlabel, x_test, label, y_test = train_test_split( x_pool, y_pool, test_size = test_size) return x_train, y_train, x_test, y_test, unlabel, labelif __name__ == '__main__': # read dataset dataset = pd.read_csv("./spambase.csv").values[:, ] # imputing missing data imputer = SimpleImputer(missing_values = 0, strategy ="mean") imputer = imputer.fit(dataset[:, :-1]) dataset[:, :-1] = imputer.transform(dataset[:, :-1]) # feature scaling sc = StandardScaler() dataset[:, :-1] = sc.fit_transform(dataset[:, :-1]) # run both models 100 times and take the average of their accuracy ac1, ac2 = [], [] # arrays to store accuracy of different models for i in range(100): # split dataset into train(5 %), test(25 %), unlabel(70 %) x_train, y_train, x_test, y_test, unlabel, label = split( dataset, 0.05, 0.25) # train model by active learning for i in range(5): classifier1 = LogisticRegression() classifier1.fit(x_train, y_train) y_probab = classifier1.predict_proba(unlabel)[:, 0] p = 0.47 # range of uncertanity 0.47 to 0.53 uncrt_pt_ind = [] for i in range(unlabel.shape[0]): if(y_probab[i] >= p and y_probab[i] <= 1-p): uncrt_pt_ind.append(i) x_train = np.append(unlabel[uncrt_pt_ind, :], x_train, axis = 0) y_train = np.append(label[uncrt_pt_ind], y_train) unlabel = np.delete(unlabel, uncrt_pt_ind, axis = 0) label = np.delete(label, uncrt_pt_ind) classifier2 = LogisticRegression() classifier2.fit(x_train, y_train) ac1.append(classifier2.score(x_test, y_test)) ''' split dataset into train(same as generated by our model), test(25 %), unlabel(rest) ''' train_size = x_train.shape[0]/dataset.shape[0] x_train, y_train, x_test, y_test, unlabel, label = split( dataset, train_size, 0.25) # train model without active learning classifier3 = LogisticRegression() classifier3.fit(x_train, y_train) ac2.append(classifier3.score(x_test, y_test)) print("Accuracy by active model :", mean(ac1)*100) print("Accuracy by random sampling :", mean(ac2)*100)'''This code is contributed by Raghav Dalmiahttps://github.com / raghav-dalmia''' |
Output:
Accuracy by active model : 80.7 Accuracy by random sampling : 79.5
There are several models for the selection of the most valuable points. Some of them are:
- Query by committee
- Query synthesis and Nearest neighbour search
- Large margin-based heuristics
- Posterior probability-based heuristics
Reference: Active Learning Synthesis Lectures on Artificial Intelligence and Machine Learning by Burr S.
