PHP Machine Learning Code:

Hey folks, till now I have discussed all the aspects related to our project but still I have not answered your question which is, Why I chose PHP over Python?

The reason being PHP is a server-side object-oriented programming language which is obviously meant for web-based application. Being server-side, its execution is very fast as compared to Python which is a scripting language. Moreover, when we are dealing with large amount of live data it is very important that the processing of data is fast and consistent which is provided by PHP. Also, PHP is able to run on many platforms but Python has its limitation on working on different platforms. Another important reason being PHP broadly supports database connectivity but on other hand Python has its limitation again when comes to database connectivity.

Here, in my project for Smart India Hackathon, which is Asset Performance Management, I proposed a Web Based Approach for implementing ML model. So, adhering to all such reasons I chose PHP over Python.

I hope you are satisfied with this explanation about why I chose PHP over Python. Again, these are my personal views as it may differ from person to person depending on project requirement.

So, let’s get started writing the ML code in PHP. Writing ML code in PHP is very easy as it takes just 5 lines of code using Rubix-ML library.

*Note – For using Rubix-ML PHP Library we need PHP Version 7.4 and later.

First to begin with, consider we need to predict the efficiency of our asset which is heat exchanger.

#Code for Predicting Efficiency of Heat Exchanger –

Okay folks, now let me explain you the code step by step…

Now every machine learning project requires some libraries and packages to work with data and to implement the algorithms. Likely, we need to import some packages/class from Rubix-ML library into our ML model.

*Note - Since, PHP is OOP language we sometimes refer packages as Class.

The libraries are:

a) include __DIR__ . '/vendor/autoload.php'; –

This is basically not a library but a Composer. Composer is an application-level package manager for the PHP programming language that provides a standard format for managing dependencies of PHP software and required libraries.

b) use Rubix\ML\Datasets\Labeled; -

Now we import the Datasets package for importing our dataset into algorithm and manipulating with our dataset. This package consists of a Class named Labeled which has predefined methods for working with the labeled dataset.

c) use Rubix\ML\Extractors\CSV; -

Although we have used the Dataset package for manipulating with the dataset but it is only possible when we import the dataset into our algorithm in a proper format. So, we need to use the Extractors package. Extractors are iterators that let you loop over the records of a dataset in storage and can be used to instantiate a dataset object using the fromIterator() method. Similarly, our dataset should be in the CSV format when using it in our ML model. Thus, Extractors package also consists of CSV Class which automatically converts dataset files into CSV format. CSV files have the advantage of being able to be processed line by line. Thus, all CSV data are imported as categorical type (strings) by default.

d) use Rubix\ML\Classifiers\KNearestNeighbors; -

Till now we have imported the dataset, now we need to import the algorithm package which we are going to implement in our code. Since, we are performing Classification in our model we need to import the Classifiers package which has the KNearestNeighbors Algorithm. So now, we have imported the Classifiers Class which means now we can apply K-NN on our dataset.

e) use Rubix\ML\Kernels\Distance\Hamming; -

After importing the Classifiers package we can now work on our algorithm but still the Distance Kernel part is unknown to our K-NN algorithm based on which it is going to calculate the nearest neighbor. So, for that we import the Kernels package which consists of the Distance package where tons of Distance metrics are available. Since we are working with the categorical data we need to import the corresponding metrics. So, we import the Hamming Distance Metric for our ML code.

f) use Rubix\ML\CrossValidation\Metrics\Accuracy; -

Now coming to the last package which is CrossValidation. Generally, CrossValidation techniques are used for evaluating ML models. Evaluating in the sense measuring the accuracy of our ML model on different parameters based on different subsets of data. Here in our ML model we are just going to measure the accuracy of our Algorithm on our provided Testing Dataset.

Now coming to the actual code let’s explore it step by step…

*Note – 1) Comments in PHP are represented using ‘//’.

2) To display output in PHP the syntax is:

echo ‘write_your_output_code/sentence_here’

3) In PHP variable is declared using ‘$’ symbol

Step 1) Loading Dataset into Algorithm:

Code:

$dataset = Labeled::fromIterator(new CSV('heat_efficiency.csv'));

Here, I have declared one variable named dataset which will store our imported dataset which is for Efficiency of Heat Exchanger. The object of the Class CSV is instantiated by using the new keyword. The new Keyword is followed by extractor named CSV(), in this extractor we need to pass the path of our required dataset and then it will convert it into CSV format. But only storing the dataset is not enough as we even want to work with the records present in that dataset. So, we need to use the Class Labeled which consists of method named fromIterator() which iterates over each record from the dataset. We can use this method fromIterator() from Class Labeled by using a scope resolution operator.

Step 2) Splitting the dataset into training and testing dataset:

Code:

[$training, $testing] = $dataset-> stratifiedSplit(0.7);

Here, again I have declared two variables named training and testing which will store the respective dataset after split, which will be operated on our imported dataset. For splitting the dataset, we use the predefined method named stratifiedSplit which takes one argument as input which is obviously the splitting factor for dataset. It takes argument values between 0 and 1. As discussed earlier, I am going to split the dataset into 70:30 ratio. Thus, I passed 0.7 as an argument to the method. Here, by default it assumes that if we are working with two variables then the splitting will be done accordingly. Thus, it splits our dataset into 70% for training and 30% for testing. That’s actually the power of Rubix-ML library.

Step 3) Applying estimator i.e. the K-NN Algorithm to the ML model:

Code:

$estimator = new KNearestNeighbors(2, true, new Hamming(2));

Hey folks, now coming to the interesting part which is applying the machine learning algorithm or you can say estimator for our ml model. Here, I have declared a variable named estimator which will be storing the ML parameters and Class i.e. Algorithm that we are going to use in our model. So, by using new keyword the object of the Class K-NN is instantiated and now we can use our algorithm. Since we are going for K-NearestNeighbors we need to use the KNearestNeighbors Class. The KNearestNeighbors have different parameters. The first parameter is the value of nearest neighbor which we will consider for our ML model. The value generally depends on the dataset we are using and the user. Here I choose 2 as the parameter of nearest neighbor since we are working on three categories of values i.e. 1) L_Efficiency, 2) M_Efficiency and 3) H_Efficiency. It literally means that whenever a new data point will be entered the K-NN algorithm will search for 2 Nearest or likely same data point/group where the entered data point suits perfectly. We cannot choose nearest parameter greater than 3 as it will cause ambiguity while the algorithm searches for likely same data point. The second parameter is the Boolean value which is set to true. It literally means that the algorithm has to consider the distances of nearest neighbors while making predictions. The third parameter which is really important is the Distance Metric. As discussed earlier, we are using the Hamming Distance Metric. So, we need to create the object of Hamming Distance Class with new keyword to instantiate it. It takes a parameter for measuring the distance. Here I have passed 2 as a parameter as we have discussed earlier in the Kernel Distance section.

Step 4) Training/Fitting the Algorithm with Training Dataset:

Code:

$estimator->train($training);

Now comes the fun part, here our model is ready to be trained. Here, we apply train method on the training variable which earlier had stored the Training Dataset. But how the algorithm is going to be trained for the dataset. For that we make the variable estimator to apply on the training variable using the train method. Variable estimator had stored the pattern building algorithm and parameters to apply on the dataset. Thus, we have trained our ML model on our Dataset for particular patterns.

Step 5) Testing/Predicting the Trained Algorithm on Testing Dataset:

Code:

$predictions = $estimator->predict($testing);

Until now, we have trained our ML model on the training dataset. Now it’s time to predict the output or test the patterns learnt by our ML model. For that I have declared another variable named predictions which will store the predicted datapoints from the testing dataset. But how it will predict? For that we need to use the predict method which will work on our testing dataset. But still the predict method don’t know how to predict. For that we need to apply the estimator on predict method as it stores the learned pattern and parameters for the ML model. Based on the learned patterns the predict method will make predictions from the testing dataset and it will store it in the variable named predictions.

Step 6) Printing the Predicted values:

Code:

$output = array_slice($predictions, 0, 5);

After we have tested/predicted the values from the testing dataset, the algorithm will store the predicted values in the form of arrays. As the algorithm is working on the complete dataset it will predict value for each group of data points. Therefore, we need to restrict or choose limited values of output for better insights. Here, I have used the array_slice method to limit the output values. It takes three parameters, first one is the predictions variable which has stored the predicted values, the second and third parameter is the starting index and ending index of the array. Since I want first five output values so I choose starting index as 0 and ending index as 5. Finally, the first 5 predicted values are stored in the variable named output.

Step 7) Finding the Accuracy of our ML Model:

Code:

$metric = new Accuracy();
      $score  =  $metric->score($predictions, $testing->labels());

Now coming into the endgame is to find how accurate our ML model worked on the testing dataset. For that we need to use the predefined Accuracy Class which will find the accuracy of our ML model. So, I instantiate the object of that class using the new keyword and stored it in the variable named metric. Now here comes the fun part, measuring accuracy of the model means the number of classifications a model correctly predicts divided by the total number of predictions made. Here, we need not do the calculations as the Accuracy class has predefined method named score which will do the complex stuff. Scoring is also called prediction, and is the process of generating values based on a trained machine learning model, given some new input data (testing data). Now, the score method takes two parameters i.e. the tested data points and the corresponding parameters based on which it has generated the result. So, I have passed predictions variable as a first parameter which holds the patterns tested on dataset and similarly the second variable named testing which holds the testing dataset, is now applied on the labels method. The labels method generally holds the output value provided by the user while training. Thus, it now cross-validates the generated data with the provided data when we apply the Accuracy class stored in variable metrics.

Step 8) Converting the Score into User format:

Code:

echo 'Accuracy of the model is ' . (string) ($score * 100.0) . '%' . PHP_EOL;

This line of code does not do any complex task but just converts the score into user format. As the score generated in the previous step was lying between 0 and 1, we convert it into percent form by multiplying it with 100 thus making it more user understandable. It fairly depends on the user whether to convert it or not.

And Voila!!! We have completed our ML Model here.

Now, you might be thinking what about the remaining datasets i.e. the Heat Exchanger Reliability and Status. Don’t worry it’s not a tedious task, you just need to replace the heat_efficiency dataset with the heat_reliability and heat_status datasets, remaining the ML model will perform the same task for both the datasets.

Until here, we have just performed the machine learning stuff where we got to know about the predictions based on the datasets provided. But this stuff is very boring as just a piece of data is predicted for an instance. Now, if we want to know how the data of heat exchanger is behaving for the complete dataset then we need to explore the dataset. Yes! You guessed it right, we are going to do Data Exploration with Visualization.

Just going off-track a bit, how many of you have watched the movie Drishyam? What does this movie convey? The movie conveys the fact that, Visual Memories are the strongest memories which helps faster communication of data with better understanding.

So, let’s have a Drishyam now (I mean Data Visualization)... 😅😅