https://github.com/boltuix/tflite-text-classifier-android-app
This Android app leverages a TensorFlow Lite model for on-device classification of social media posts into 11 categories, including technology, sports, finance, and more. Built with Kotlin and Jetpack Compose, it ensures efficient, privacy-focused inference without server dependencies.
https://github.com/boltuix/tflite-text-classifier-android-app
ai android appdev colab-notebook jetpack-compose jetpackcompose kotlin machine-learning mobile-app nlp on-device tensorflow tensorflow-lite text-classification tflite tflite-models tips-and-tricks
Last synced: 5 months ago
JSON representation
This Android app leverages a TensorFlow Lite model for on-device classification of social media posts into 11 categories, including technology, sports, finance, and more. Built with Kotlin and Jetpack Compose, it ensures efficient, privacy-focused inference without server dependencies.
- Host: GitHub
- URL: https://github.com/boltuix/tflite-text-classifier-android-app
- Owner: BoltUIX
- Created: 2025-08-26T03:21:03.000Z (6 months ago)
- Default Branch: main
- Last Pushed: 2025-08-26T04:27:01.000Z (6 months ago)
- Last Synced: 2025-08-26T05:32:58.314Z (6 months ago)
- Topics: ai, android, appdev, colab-notebook, jetpack-compose, jetpackcompose, kotlin, machine-learning, mobile-app, nlp, on-device, tensorflow, tensorflow-lite, text-classification, tflite, tflite-models, tips-and-tricks
- Language: Jupyter Notebook
- Homepage: https://www.boltuix.com/2025/08/how-to-use-tensorflow-lite-for-text.html
- Size: 1.6 MB
- Stars: 1
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TFLite Text Classifier Android App
------------------------------------------------------------------------
This Android app leverages a TensorFlow Lite model for on-device
classification of social media posts into 10+ categories, including
technology, sports, finance, and more. Built with Kotlin and Jetpack
Compose, it ensures efficient, privacy-focused inference without server
dependencies.
[](https://opensource.org/licenses/MIT)
[](https://www.tensorflow.org/lite)
[](https://kotlinlang.org/)
[](https://developer.android.com/jetpack/compose)
[](https://developer.android.com/)
------------------------------------------------------------------------
## Features
- **On-device classification** using TensorFlow Lite
- **Jetpack Compose UI** for modern Android apps
- **Multi-line input** for flexible text entry
- **Probability distribution** for each category
- **Fast & lightweight inference** with no server calls
- **Manual tokenizer** with `vocab.txt` for reproducibility
------------------------------------------------------------------------
## Training Notebook -- `model.ipynb`
The repository includes a Jupyter notebook **`Android.ipynb`**,
containing the full **training pipeline** for the classifier:
- Load & preprocess training data
- Tokenization with `TextVectorization`
- Build & train a Keras model
- Convert the model to `.tflite` format
- Export `vocab.txt` and `labels.txt` for Android usage
Use this notebook to retrain or fine-tune your model while ensuring
**consistency** between training and deployment.
------------------------------------------------------------------------
## Model Overview
- **Model file**: `model_with_softmax.tflite`
- **Input**: `int32[1, 50]` → tokenized & padded sequence
- **Output**: `float32[1, 10]` → probability vector (softmax)
- **Tokenizer**: `vocab.txt` from training (`TextVectorization`)
- **Labels**: `labels.txt` with 11 categories
------------------------------------------------------------------------
## Folder Structure
app/
├── assets/
│ ├── model_with_softmax.tflite
│ ├── vocab.txt
│ └── labels.txt
└── res/
└── layout/ (if legacy XML UI is present)
------------------------------------------------------------------------
## Setup Instructions
1. Add dependencies in `build.gradle`:
``` gradle
implementation 'org.tensorflow:tensorflow-lite:2.12.0'
implementation 'org.tensorflow:tensorflow-lite-support:0.3.1'
```
2. Copy required files to `app/src/main/assets/`:
- `model_with_softmax.tflite`
- `vocab.txt`
- `labels.txt`
3. Initialize the classifier in Kotlin:
``` kotlin
TextClassifier.initialize(context)
```
------------------------------------------------------------------------
## Example Usage (Jetpack Compose)
``` kotlin
val text by remember { mutableStateOf("") }
Column {
TextField(
value = text,
onValueChange = { text = it },
placeholder = { Text("Enter your post...") },
modifier = Modifier
.fillMaxWidth()
.height(150.dp),
maxLines = 5
)
Button(onClick = {
val result = TextClassifier.predictAll(text)
val formatted = result.joinToString("\n") { (label, prob) ->
"$label: ${"%.2f".format(prob * 100)}%"
}
Log.d("Prediction", formatted)
}) {
Text("Classify")
}
}
```
------------------------------------------------------------------------
## Notes
- The `.tflite` model **must include softmax**
- `vocab.txt` and `labels.txt` must match training data
- Assets must be placed in `app/src/main/assets/` (not `res/`)
- Use `coerceIn()` to ensure predictions remain between `0–1`
------------------------------------------------------------------------
## License
MIT --- Free to use, modify, and distribute.