Upload app.py with huggingface_hub
Browse files
app.py
ADDED
|
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Aplikasi Gradio — Prediksi Segmen Pelanggan E-commerce
|
| 3 |
+
UAS Business Data Science (MN1135)
|
| 4 |
+
|
| 5 |
+
Memuat model Decision Tree hasil tuning beserta objek pra-pemrosesannya, lalu
|
| 6 |
+
menyediakan antarmuka web agar siapa pun dapat mencoba model tanpa membuka notebook.
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
import joblib
|
| 10 |
+
import numpy as np
|
| 11 |
+
import pandas as pd
|
| 12 |
+
import gradio as gr
|
| 13 |
+
|
| 14 |
+
# ---------------------------------------------------------------- muat artefak
|
| 15 |
+
model = joblib.load("model_segmen.joblib")
|
| 16 |
+
scaler = joblib.load("scaler.joblib")
|
| 17 |
+
FITUR = joblib.load("feature_names.joblib") # urutan kolom saat pelatihan
|
| 18 |
+
META = joblib.load("metadata.joblib")
|
| 19 |
+
|
| 20 |
+
NUMERIK = META["kolom_numerik"]
|
| 21 |
+
KATEGORIKAL = META["kolom_kategorikal"]
|
| 22 |
+
SKALA = META["kolom_numerik_scaling"]
|
| 23 |
+
PERLU_SCALING = META["perlu_scaling"]
|
| 24 |
+
LABEL = META["label_kelas"]
|
| 25 |
+
PILIHAN = META["pilihan_kategori"]
|
| 26 |
+
RENTANG = META["rentang_input"]
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
def siapkan_baris(data_mentah: dict) -> pd.DataFrame:
|
| 30 |
+
"""Ubah satu baris input mentah menjadi baris fitur yang dikenali model.
|
| 31 |
+
|
| 32 |
+
Langkahnya harus SAMA PERSIS dengan pra-pemrosesan di notebook: one-hot encoding
|
| 33 |
+
dengan drop_first, urutan kolom mengikuti FITUR, dan scaling hanya bila model
|
| 34 |
+
dilatih pada data ter-scaling.
|
| 35 |
+
|
| 36 |
+
Catatan penting: one-hot TIDAK boleh dibuat dengan pd.get_dummies(). Pada satu baris,
|
| 37 |
+
setiap kolom kategorikal hanya punya satu nilai, sehingga drop_first membuang kolom itu
|
| 38 |
+
dan hasilnya salah tanpa pesan galat. Karena itu kolom dummy dinyalakan satu per satu.
|
| 39 |
+
"""
|
| 40 |
+
baris = pd.DataFrame(np.zeros((1, len(FITUR))), columns=FITUR)
|
| 41 |
+
|
| 42 |
+
for kolom in NUMERIK:
|
| 43 |
+
baris.at[0, kolom] = data_mentah[kolom]
|
| 44 |
+
|
| 45 |
+
for kolom in KATEGORIKAL:
|
| 46 |
+
nama_dummy = f"{kolom}_{data_mentah[kolom]}"
|
| 47 |
+
if nama_dummy in baris.columns:
|
| 48 |
+
baris.at[0, nama_dummy] = 1
|
| 49 |
+
|
| 50 |
+
if PERLU_SCALING:
|
| 51 |
+
baris[SKALA] = scaler.transform(baris[SKALA])
|
| 52 |
+
return baris
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
def prediksi(umur, gender, kota, negara_bagian, negara, region, kode_pos, biaya_akuisisi):
|
| 56 |
+
data_mentah = {
|
| 57 |
+
"customer_age": umur,
|
| 58 |
+
"gender": gender,
|
| 59 |
+
"customer_city": str(kota).strip(),
|
| 60 |
+
"customer_state": negara_bagian,
|
| 61 |
+
"customer_country": negara,
|
| 62 |
+
"region": region,
|
| 63 |
+
"customer_postal_code": kode_pos,
|
| 64 |
+
"customer_acquisition_cost": biaya_akuisisi,
|
| 65 |
+
}
|
| 66 |
+
baris = siapkan_baris(data_mentah)
|
| 67 |
+
proba = model.predict_proba(baris.to_numpy(dtype=np.float32))[0]
|
| 68 |
+
|
| 69 |
+
hasil = {LABEL[str(int(k))]: float(p) for k, p in zip(model.classes_, proba)}
|
| 70 |
+
segmen = max(hasil, key=hasil.get)
|
| 71 |
+
catatan = (f"Segmen yang diprediksi: {segmen}. "
|
| 72 |
+
"Perhatian: pada dataset ini performa model setara tebakan acak "
|
| 73 |
+
"(F1 macro sekitar 0,25), sehingga hasil ini hanya demonstrasi alur deployment.")
|
| 74 |
+
return hasil, catatan
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
# ------------------------------------------------------------------ antarmuka
|
| 78 |
+
antarmuka = gr.Interface(
|
| 79 |
+
fn=prediksi,
|
| 80 |
+
inputs=[
|
| 81 |
+
gr.Slider(RENTANG["customer_age"][0], RENTANG["customer_age"][1], value=45, step=1,
|
| 82 |
+
label="Umur pelanggan (tahun)"),
|
| 83 |
+
gr.Radio(PILIHAN["gender"], value=PILIHAN["gender"][0], label="Gender"),
|
| 84 |
+
gr.Textbox(value="Laurenland", label="Kota",
|
| 85 |
+
info="Ketik nama kota. Kota yang tidak ada di data latih tetap diproses (semua dummy kota = 0)."),
|
| 86 |
+
gr.Dropdown(PILIHAN["customer_state"], value=PILIHAN["customer_state"][0], label="Negara bagian / provinsi"),
|
| 87 |
+
gr.Dropdown(PILIHAN["customer_country"], value=PILIHAN["customer_country"][0], label="Negara"),
|
| 88 |
+
gr.Radio(PILIHAN["region"], value=PILIHAN["region"][0], label="Region"),
|
| 89 |
+
gr.Number(value=11253, precision=0, label="Kode pos"),
|
| 90 |
+
gr.Number(value=30.0, label="Biaya akuisisi pelanggan"),
|
| 91 |
+
],
|
| 92 |
+
outputs=[
|
| 93 |
+
gr.Label(num_top_classes=4, label="Peluang tiap segmen"),
|
| 94 |
+
gr.Textbox(label="Keterangan"),
|
| 95 |
+
],
|
| 96 |
+
title="Prediksi Segmen Pelanggan E-commerce",
|
| 97 |
+
description=(
|
| 98 |
+
f"Model: {META['nama_model']} · F1 macro pada data uji: {META['metrik_data_uji']['f1_macro']} · "
|
| 99 |
+
f"Accuracy: {META['metrik_data_uji']['accuracy']}. "
|
| 100 |
+
"Isi profil pelanggan, lalu tekan Submit."
|
| 101 |
+
),
|
| 102 |
+
article=(
|
| 103 |
+
"**Keterbatasan.** Fitur yang tersedia (umur, biaya akuisisi, gender, lokasi) hampir tidak membedakan segmen, "
|
| 104 |
+
"sehingga performa model setara tebakan acak. Jangan dipakai untuk keputusan pemasaran sungguhan. "
|
| 105 |
+
"Peluang bernilai 0 atau 1 karena pohon dibiarkan tumbuh penuh (tanpa batas kedalaman)."
|
| 106 |
+
),
|
| 107 |
+
flagging_mode="never",
|
| 108 |
+
)
|
| 109 |
+
|
| 110 |
+
if __name__ == "__main__":
|
| 111 |
+
antarmuka.launch(share=True) # share=True menghasilkan tautan publik sementara
|