knn in parallel with julia

K-Nearest Neighbors Algorithm in Parallel using Julia

K-Nearest Neighbors algorithm known of being computationally intensive because the distance to every point has to be calculated.

Here we're going to use Julia's parallel computing power to fit KNN algorithm on Census Income dataset to predict whether income exceeds $50K ...

Data can be downloaded from https://archive.ics.uci.edu/ml/machine-learning-databases/adult/ ; they're two datasets, training and test.

We will be using Julia version 0.6.4

# run the script from the command line
$ julia knn_income.jl

### READING DATA ###
(:data_shape, (32561, 15), (16281, 15))

# Preparing Data:
1 Deleting any row containing any unknown value.
(:new_data_shape, (30162, 15), (15060, 15))

2 one hot encoding string columns and normalizing numeric ones.
(:encoded_data_shape, (30162, 105), (15060, 105))

3 Splitting data into train and test
# size of train and test data
(104, 30162)(104, 15060)

# Define the Model:
* Adding 4 processes => define euclidean distance function
        => get k nearest neighbors => assign new labels
(:number_of_processes, 5)

# Get predictions and accuracy on train and test data using 40 k
(:train_accuracy, 0.8310456866255553)

(:test_accuracy, 0.8301460823373173)