gudbjartur22/clinx02-androidpebble
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

V4.2.1a - added standard deviation threshold to ML Algorithm

Browse Source
This commit is contained in:
Graham Jones
2023-08-19 19:54:19 +01:00
parent 70f7a46764
commit bc455a2f7a
6 changed files with 191 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

123
app/src/main/java/uk/org/openseizuredetector/SdAlgNn.java
View File

@@ -1,7 +1,10 @@
package uk.org.openseizuredetector;
import android.content.Context;
import android.content.SharedPreferences;
import android.preference.PreferenceManager;
import android.util.Log;
import android.widget.Toast;
import com.android.volley.AuthFailureError;
import com.android.volley.Request;
@@ -31,28 +34,50 @@ public class SdAlgNn {
private Context mContext;
RequestQueue mQueue;
private double mSdThresh; // Acceleration Standard Deviation Threshold required to activate analysis (%)
private int mModelId; // ID of ML Model to be used (refers to information in MlModels.json for details).
private int mInputFormat; // ID of input format required for model (populated from MlModels.json).
public SdAlgNn(Context context) {
Log.d(TAG, "SdAlgNn Constructor");
mContext = context;
SharedPreferences SP = PreferenceManager
.getDefaultSharedPreferences(mContext);
try {
String threshStr = SP.getString("CnnAlarmThreshold", "5");
mSdThresh = Double.parseDouble(threshStr);
Log.v(TAG, "SdAlgNn Constructor mSdThresh = " + mSdThresh);
threshStr = SP.getString("CnnModelId", "1");
mModelId = Integer.parseInt(threshStr);
Log.v(TAG, "SdAlgNn Constructor mModelId = " + mModelId);
} catch (Exception ex) {
Log.v(TAG, "SdAlgNn Constructor - problem parsing preferences. " + ex.toString());
Toast toast = Toast.makeText(mContext, "Problem Parsing ML Algorithm Preferences", Toast.LENGTH_SHORT);
toast.show();
}
mInputFormat = 1; // FIXME - this needs to be determined from the model ID specified by retrieving a configuration file.
Task<Void> initializeTask = TfLite.initialize(mContext);
initializeTask.addOnSuccessListener(a -> {
MappedByteBuffer modelBuffer;
try {
Log.d(TAG, "onSuccessListener - loading model");
modelBuffer = FileUtil.loadMappedFile(context, MODEL_PATH);
Log.d(TAG, "onSuccessListener - model loaded");
} catch (IOException e) {
Log.e(TAG, "Error Loading Model File");
return;
}
Log.d(TAG, "onSuccessListener - creating interpreter");
interpreter = InterpreterApi.create(modelBuffer,
new InterpreterApi.Options().setRuntime(
InterpreterApi.Options.TfLiteRuntime.FROM_SYSTEM_ONLY));
Log.d(TAG, "onSuccessListener - interpreter created ok");
})
MappedByteBuffer modelBuffer;
try {
Log.d(TAG, "onSuccessListener - loading model");
modelBuffer = FileUtil.loadMappedFile(context, MODEL_PATH);
Log.d(TAG, "onSuccessListener - model loaded");
} catch (IOException e) {
Log.e(TAG, "Error Loading Model File");
return;
}
Log.d(TAG, "onSuccessListener - creating interpreter");
interpreter = InterpreterApi.create(modelBuffer,
new InterpreterApi.Options().setRuntime(
InterpreterApi.Options.TfLiteRuntime.FROM_SYSTEM_ONLY));
Log.d(TAG, "onSuccessListener - interpreter created ok");
})
.addOnFailureListener(e -> {
Log.e(TAG, String.format("Cannot initialize interpreter: %s",
e.getMessage()));
@@ -62,25 +87,81 @@ public class SdAlgNn {
}
public void close() {
Log.d(TAG,"close()");
Log.d(TAG, "close()");
if (interpreter != null) {
interpreter.close();
}
}
public float getPseizure(SdData sdData) {
/**
* getPseizureFmt1 - calculate probability of sdData representing seizure-like movement
* using a model with input format #1, which is a simple vector of 125 accelerometer vector
* magnitude readings.
* @param sdData - seizure detector data as input to the model
* @return probability of data representing seizure-like movement.
*/
private float getPseizureFmt1(SdData sdData) {
int i;
float[][][] modelInput = new float[1][125][1];
float[][] modelOutput = new float[1][2];
for (int j = 0; j < 125; j++) {
modelInput[0][j][0] = (float)sdData.rawData[j];
modelInput[0][j][0] = (float) sdData.rawData[j];
}
if (interpreter == null) {
Log.d(TAG,"getPSeizure() - interpreter is null - returning zero seizure probability");
Log.d(TAG, "getPSeizure() - interpreter is null - returning zero seizure probability");
return (0.0f);
}
interpreter.run(modelInput, modelOutput);
Log.d(TAG,"run - pSeizure="+modelOutput[0][1]);
return(modelOutput[0][1]);
Log.d(TAG, "run - pSeizure=" + modelOutput[0][1]);
return (modelOutput[0][1]);
}
public float getPseizure(SdData sdData) {
int i;
// First check that we have enough movement to analyse by comparing the acceleration standard deviation to a threshold.
double stdDev;
stdDev = calcRawDataStd(sdData);
if (stdDev < mSdThresh) {
Log.d(TAG, "getPseizure - acceleration stdev below movement threshold: std="+stdDev+", thresh="+mSdThresh);
return (0);
}
float pSeizure;
switch (mModelId) {
case 1:
pSeizure = getPseizureFmt1(sdData);
break;
default:
Log.e(TAG,"getPSeizure - invalid model ID "+mModelId);
pSeizure = 0;
}
return(pSeizure);
}
private double calcRawDataStd(SdData sdData) {
/**
* Calculate the standard deviation in % of the rawData array in the SdData instance provided.
* It assumes that rawdata will contain 125 samples.
* Returns the standard deviation in %.
*/
// FIXME - assumes length of rawdata array is 125 data points
int j;
double sum = 0.0;
for (j = 0; j < 125; j++) { // FIXME - assumed length!
sum += sdData.rawData[j];
}
double mean = sum / 125;
double standardDeviation = 0.0;
for (j = 0; j < 125; j++) { // FIXME - assumed length!
standardDeviation += Math.pow(sdData.rawData[j] - mean, 2);
}
standardDeviation = Math.sqrt(standardDeviation / 125); // FIXME - assumed length!
// Convert standard deviation from milli-g to %
standardDeviation = 100. * standardDeviation / mean;
return (standardDeviation);
}
}