Meiosis Phase Classifier (KDS)

Classify meiotic phases from microscopy images using OpenCV preprocessing, spatial feature extraction, and a Random Forest model.

Repository: github.com/valereoo/KDS

---

How to run this repo

Goal	Where	How
Train / regenerate model	Google Colab	Open notebook in Colab
Use the classifier (UI)	Local or Streamlit Cloud	Run the Streamlit app

Live app: kdskelompok17.streamlit.app

---

Project structure

KDS/
├── notebooks/             # Training notebook (run on Colab)
├── data/                  # CSV inputs/outputs (gitignored if large)
├── models/                # Trained .joblib for the web app
├── app.py                 # Streamlit entry point
├── src/kds/               # App logic used by Streamlit
└── requirements.txt

---

1. Train the model (Google Colab) — recommended

The training pipeline lives in:

notebooks/Kelompok17_K04_IF3211_18223010_18223016_18223030_18223068_klasifikasi_fase_meiosis_opencv.ipynb

It downloads the Kaggle dataset hasan1101/mitosis-detection, extracts OpenCV features, trains a Random Forest, and writes artifacts to outputs_meiosis_opencv/.

Open in Colab (one click)

Or clone inside a new Colab notebook

Run this in the first cell of Colab, then open the notebook from the file browser (notebooks/ → your .ipynb):

!git clone https://github.com/valereoo/KDS.git
%cd KDS/notebooks

Colab setup checklist

1. Runtime → Change runtime type → Python 3 (GPU not required).
2. Run cells top to bottom from the training notebook.
3. Kaggle data — the notebook uses kagglehub to download the dataset automatically. If download fails:
   • Create a Kaggle API token (kaggle.json).
   • In Colab: Upload kaggle.json, then run:

     !mkdir -p ~/.kaggle && mv kaggle.json ~/.kaggle/ && chmod 600 ~/.kaggle/kaggle.json

   • Re-run the dataset download cell.
4. Optional manual labels — place manual_phase_labels.csv in the notebook working directory (notebooks/ when using %cd KDS/notebooks), or use the template the notebook generates.

After training on Colab

The notebook saves files under outputs_meiosis_opencv/. Copy the important ones into the repo layout so the Streamlit app can use them:

From (Colab)	Copy to (repo)
outputs_meiosis_opencv/meiosis_phase_opencv_random_forest.joblib	models/
outputs_meiosis_opencv/opencv_spatial_features.csv	data/
outputs_meiosis_opencv/manual_phase_labels_template.csv	data/
outputs_meiosis_opencv/predictions_test.csv	data/
outputs_meiosis_opencv/feature_importance.csv	data/
outputs_meiosis_opencv/permutation_importance.csv	data/

Download from Colab (run in a new cell after training):

from google.colab import files
from pathlib import Path
import shutil

out = Path("outputs_meiosis_opencv")
zip_path = "/content/kds_artifacts.zip"
shutil.make_archive("/content/kds_artifacts", "zip", out)
files.download(zip_path)

Unzip locally into models/ and data/ as in the table above.

Commit the model to GitHub (use Git LFS for the .joblib):

git lfs install
git lfs track "*.joblib"
git add models/meiosis_phase_opencv_random_forest.joblib
git commit -m "Add trained model"
git push

---

2. Run the web app (Streamlit)

The interactive classifier is app.py. It needs the trained model in models/meiosis_phase_opencv_random_forest.joblib (from Colab or a previous local run).

Local

git clone https://github.com/valereoo/KDS.git
cd KDS

python -m venv .venv
.venv\Scripts\activate        # Windows
# source .venv/bin/activate   # macOS / Linux

pip install -r requirements.txt
streamlit run app.py

Open the URL shown in the terminal (usually http://localhost:8501).

Streamlit Community Cloud (deploy)

1. Push this repo to GitHub (include the model via Git LFS, or upload it after first Colab run).
2. Go to share.streamlit.io → sign in with GitHub.
3. New app → repository valereoo/KDS → Main file: app.py → Deploy.

Deployed app: https://kdskelompok17.streamlit.app/

---

3. Clone and push to GitHub

git clone https://github.com/valereoo/KDS.git
cd KDS

git lfs install
git lfs track "*.joblib"

git add .
git commit -m "Your message"
git push origin main

• env/, .venv/, and large *.csv files in data/ are gitignored by default.
• Track .joblib with Git LFS (see models/README.md).

---

Where to put files

File	Folder
Training notebook (.ipynb)	notebooks/
CSV datasets & exports	data/
Trained model (.joblib)	models/
Virtual environment	env/ or .venv/ (local only, not committed)

---

Environment variables

Variable	Description
KDS_MODEL_PATH	Override path to the .joblib model

---

License

Add your license here (e.g. MIT).