shubham0204

# Channel Mixing MLPs : Allow communication within channels ( features of embeddings )
def channel_mixing( x , channel_mixing_mlp_dims ):
    # x is a tensor of shape ( batch_size , num_patches , channels )
    x = tf.keras.layers.LayerNormalization( epsilon=1e-6 )( x )
    x = mlp( x , channel_mixing_mlp_dims )
    return x

shubham0204

# Token Mixing MLPs : Allow communication within patches.
def token_mixing( x , token_mixing_mlp_dims ):
    # x is a tensor of shape ( batch_size , num_patches , channels )
    x = tf.keras.layers.LayerNormalization( epsilon=1e-6 )( x )
    x = tf.keras.layers.Permute( dims=[ 2 , 1 ] )( x ) 
    # After transposition, shape of x -> ( batch_size , channels , num_patches )
    x = mlp( x , token_mixing_mlp_dims )
    return x

shubham0204

# Multilayer Perceptron with GeLU ( Gaussian Linear Units ) activation
def mlp( x , hidden_dims ):
    y = tf.keras.layers.Dense( hidden_dims )( x )
    y = tf.nn.gelu( y )
    y = tf.keras.layers.Dense( x.shape[ -1 ] )( y )
    y = tf.keras.layers.Dropout( 0.4 )( y )
    return y

shubham0204

// Launch a coroutine
coroutineScope.launch {
    
    results = runOnCoroutine( my_data )
    
    // You can update the UI as well!
    
    nameTextView.text = results.userName
    ageTextView.text = results.userAge.toString()
    

shubham0204

private suspend fun runOnCoroutine( some_data : Data ) = 
    withContext( Dispatchers.Default ) { 
        results = my_lazy_function(data)
        return@withContext results 
    }

shubham0204

...

// Called when predict_button is clicked. ( See activity_main.xml ).
fun onPredictButtonClick( view : View ) {
    // Split the String by ","
    var strX = sampleInputEditText.text.toString().split( "," ).toTypedArray()
    strX = strX.map{ xi -> xi.trim() }.toTypedArray()
    // Convert the String[] to float[]
    val x = dataFrame.convertStringArrayToFloatArray( strX )
    // Predict the class with GaussianNB.

shubham0204

class MainActivity : AppCompatActivity() {

    // View elements to take the inputs and display it to the user.
    private lateinit var sampleInputEditText : TextInputEditText
    private lateinit var outputTextView : TextView

    // DataFrame object which will hold the data.
    private lateinit var dataFrame: DataFrame

    // GaussianNB object to perform the calculations and return

shubham0204

// Predict the label for the given sample.
private fun predictLabel( sample : FloatArray ) {
    val probArray = FloatArray( dataFrame.numClasses )
    for ( ( i , priorProb ) in priorProbabilities.values.withIndex()) {
        // We take the log probabilities so as to avoid underflow.
        var p = log10( priorProb )
        // While we take log, the product is transformed into a sum
        // log( a . b ) = log(a) + log(b)
        for ( j in 0 until dataFrame.numFeatures ) {
            p += featureDistributions[ j ].getLogProb( sample[ i ] )

shubham0204

// Class to implement Gaussian Naive Bayes
class GaussianNB( private var dataFrame : DataFrame ) {

    // Array to store Gaussian Distributions for each feature.
    private var featureDistributions : Array<GaussianDistribution>

    // Prior probabilities stored in a HashMap of form ( column_name , prior_prob )
    private var priorProbabilities : HashMap<String,Float>

    private var resultCallback : ResultCallback? = null

shubham0204

// A Gaussian Distribution which will be constructed for every feature, given the mean and standard deviation.
class GaussianDistribution( var mean : Float , var stdDev : Float ) {

    private val p = 1f / ( sqrt( 2.0 * Math.PI ) ).toFloat()

    // Get the likelihood for given x.
    fun getProb( x : Float ) : Float {
        val exp = exp( -0.5 * ((x - mean) / stdDev ).pow( 2 ) ).toFloat()
        return ( 1 / stdDev ) * p * exp
    }