TIF_E41212325/app.py

272 lines
10 KiB
Python

import streamlit as st
import joblib
import pandas as pd
import numpy as np
from sklearn.preprocessing import MinMaxScaler
from tensorflow.keras.models import load_model
from datetime import date, timedelta
import yfinance as yf
from pathlib import Path
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from matplotlib import rcParams
# Set font
rcParams['font.family'] = 'DejaVu Sans'
# --- Konfigurasi Halaman ---
st.set_page_config(page_title="Prediksi Harga Ethereum (GRU)", page_icon="🪙")
# --- GLOBAL STYLES ---
st.markdown("""
<style>
.app-title {
text-align:center; font-weight:800; font-size: 32px;
background: linear-gradient(90deg, #6EE7F9 0%, #7C3AED 50%, #F59E0B 100%);
-webkit-background-clip: text; -webkit-text-fill-color: transparent;
margin: 0.25rem 0 0.5rem 0;
}
.app-subtitle { text-align:center; color: #6b7280; margin-bottom: 1.25rem; }
.card {
background: #ffffff; border: 1px solid #e5e7eb; border-radius: 16px;
padding: 1rem 1.25rem; box-shadow: 0 6px 20px rgba(0,0,0,0.05); margin-bottom: 1rem;
}
.h-section { font-weight:700; font-size: 20px; margin: 0 0 .5rem 0; }
.footer { text-align:center; color: #6b7280; font-size: 13px; margin-top: 2rem; }
</style>
""", unsafe_allow_html=True)
# --- TOMBOL RESET CACHE (PENTING!) ---
with st.sidebar:
st.header("⚙️ Kontrol Data")
if st.button("🔄 Paksa Update Data (Clear Cache)"):
st.cache_data.clear()
st.success("Cache dihapus! Silakan tekan 'R' untuk reload.")
st.stop() # Hentikan app sebentar biar user reload
# --- FUNGSI LOAD DATA (VERSI YFINANCE ONLY) ---
@st.cache_data(ttl="1h")
def load_eth_data():
"""
Fokus: Download via Library yfinance.
Fitur:
1. Mengatasi data bolong (Resample Daily).
2. Auto Adjust harga (OHLC bersih).
"""
ticker = "ETH-USD"
# Kita set start date agak jauh biar grafiknya bagus
start_date = "2024-01-01"
end_date = date.today() + timedelta(days=1)
st.toast("Sedang menghubungi server Yahoo Finance...", icon="")
try:
# DOWNLOAD ONLINE
df = yf.download(
ticker,
start=start_date,
end=end_date,
progress=False,
auto_adjust=True, # Biar harga bersih
multi_level_index=False
)
if df is not None and not df.empty:
# 1. Bersihkan Index
df = df.reset_index()
# 2. Rapikan Kolom (Cegah MultiIndex/Tuple)
new_cols = []
for col in df.columns:
col_name = col[0] if isinstance(col, tuple) else str(col)
new_cols.append(col_name)
df.columns = new_cols
# 3. Pastikan kolom Date ada
if 'Date' not in df.columns:
df = df.rename(columns={df.columns[0]: 'Date'})
# 4. Hapus Timezone
df['Date'] = pd.to_datetime(df['Date']).dt.tz_localize(None)
# --- BAGIAN PENTING: TAMBAL DATA BOLONG (RESAMPLING) ---
# Ini mengatasi masalah "Loncat" dari tgl 15 ke 17.
# Kita paksa buat tanggal harian (Daily) lengkap.
df = df.sort_values('Date').set_index('Date')
# Buat range tanggal penuh dari awal sampai akhir data
full_idx = pd.date_range(start=df.index.min(), end=df.index.max(), freq='D')
# Reindex & Forward Fill (Isi kekosongan dengan data hari sebelumnya)
df = df.reindex(full_idx).ffill().reset_index()
df = df.rename(columns={'index': 'Date'})
# -------------------------------------------------------
# 5. Simpan Backup Otomatis
try:
df.to_csv("eth_backup.csv", index=False)
except:
pass
return df, "online"
except Exception as e:
print(f"Error yfinance: {e}")
# FALLBACK: JIKA DOWNLOAD GAGAL, BACA BACKUP LAMA
try:
df_backup = pd.read_csv("eth_backup.csv")
if "Unnamed: 0" in df_backup.columns: df_backup = df_backup.drop(columns=["Unnamed: 0"])
if "Date" not in df_backup.columns: df_backup = df_backup.rename(columns={df_backup.columns[0]: "Date"})
df_backup["Date"] = pd.to_datetime(df_backup["Date"]).dt.tz_localize(None)
return df_backup, "backup"
except:
return None, "error"
# --- Fungsi Helper Lainnya (Model & Scaler) ---
def validate_scaler(scaler):
issues, warnings = [], []
if not isinstance(scaler, MinMaxScaler): warnings.append(f"Scaler bukan MinMaxScaler.")
return (len(issues)==0), issues, warnings
def validate_model_input(model):
issues = []
try:
shape = model.input_shape
if shape[2] != 1: issues.append(f"Fitur model = {shape[2]}, input = 1.")
except: pass
return (len(issues)==0), issues
@st.cache_resource
def load_gru_assets():
model_path = Path("gru_model.h5")
scaler_path = Path("scaler_gru.pkl")
if not model_path.exists() or not scaler_path.exists(): return None, None
try:
model = load_model(model_path, compile=False)
scaler = joblib.load(scaler_path)
return model, scaler
except: return None, None
def predict_from_sequence_pure(model, initial_sequence_scaled, n_days):
seq = np.array(initial_sequence_scaled, dtype="float32").reshape(-1)
preds = []
for _ in range(n_days):
x = seq.reshape(1, -1, 1)
y = float(model.predict(x, verbose=0)[0, 0])
preds.append(y)
seq = np.append(seq[1:], y)
return np.array(preds, dtype="float32").reshape(-1, 1)
def create_combined_chart(df, start_date, future_dates, future_predictions):
context_start = start_date - timedelta(days=60)
recent_df = df[(df["Date"] >= context_start) & (df["Date"] <= start_date)].copy()
recent_df["Trend"] = recent_df["Close"].rolling(window=7).mean()
fig, ax = plt.subplots(figsize=(14, 8))
ax.plot(recent_df["Date"], recent_df["Close"], color="gray", alpha=0.5, label="Harga Aktual")
ax.plot(recent_df["Date"], recent_df["Trend"], color="#1f77b4", linewidth=2, label="Tren Aktual")
ax.plot(future_dates, future_predictions.flatten(), color="#d62728", linewidth=2, marker="o", markersize=4, label="Prediksi GRU")
if not recent_df.empty:
ax.plot([recent_df["Date"].iloc[-1], future_dates[0]],
[recent_df["Trend"].dropna().iloc[-1], future_predictions.flatten()[0]],
color="#d62728", linestyle="--", alpha=0.7)
ax.set_title("Analisis Tren & Prediksi Ethereum")
ax.xaxis.set_major_formatter(mdates.DateFormatter("%b %d, %Y"))
plt.xticks(rotation=45)
ax.legend()
ax.grid(True, alpha=0.3)
return fig
# --- MAIN UI ---
st.markdown("<div class='app-title'>Prediksi Harga Ethereum (GRU)</div>", unsafe_allow_html=True)
st.markdown("<div class='app-subtitle'>Data historis & prediksi ETH-USD berbasis GRU</div>", unsafe_allow_html=True)
# LOAD DATA
df, status = load_eth_data()
# Notifikasi Status
if status == "backup":
st.toast("Koneksi Yahoo Gagal. Pakai Backup.", icon="⚠️")
elif status == "error":
st.error("❌ Gagal memuat data (Library gagal & Backup tidak ada).")
st.stop()
# LOAD MODEL
model, scaler = load_gru_assets()
if df is not None and model is not None and scaler is not None:
# INFO UPDATE DATA
last_date = df["Date"].max()
st.info(f"📅 Data Terupdate sampai: **{last_date.strftime('%d %B %Y')}**")
# 1. Tampilkan Data
st.markdown("<div class='card'>", unsafe_allow_html=True)
st.markdown("<div class='h-section'>📚 Data Historis Harga Ethereum</div>", unsafe_allow_html=True)
st.dataframe(df.tail(10).sort_values("Date", ascending=False), height=300, use_container_width=True) # Tampilkan 10 data terakhir biar kelihatan update
st.markdown("</div>", unsafe_allow_html=True)
# 2. Chart Historis
st.markdown("<div class='card'>", unsafe_allow_html=True)
st.markdown("<div class='h-section'>📈 Visualisasi Harga Historis</div>", unsafe_allow_html=True)
st.line_chart(df.set_index("Date")["Close"])
st.markdown("</div>", unsafe_allow_html=True)
# 3. Prediksi
st.markdown("<div class='card'>", unsafe_allow_html=True)
st.markdown("<div class='h-section'>🎯 Mulai Prediksi</div>", unsafe_allow_html=True)
time_step = int(model.input_shape[1])
min_date = (df["Date"].min() + timedelta(days=time_step)).date()
max_date = df["Date"].max().date()
c1, c2 = st.columns(2)
with c1:
# Default value ke hari ini (max_date)
start_date = st.date_input("Mulai tanggal:", value=max_date, min_value=min_date, max_value=max_date)
with c2:
days = st.slider("Jumlah hari:", 1, 30, 15)
if st.button("Buat Prediksi"):
with st.spinner("Memproses..."):
vals = df[["Close"]].values.astype(float)
scaled = scaler.transform(vals)
check_date = pd.to_datetime(start_date)
# Cari index tanggal
idx = df[df["Date"].dt.date == check_date.date()].index
if len(idx) > 0:
idx = idx[0]
if idx < time_step:
st.error(f"Data tidak cukup (butuh {time_step} hari sebelumnya).")
else:
seq = scaled[idx-time_step : idx]
pred_scaled = predict_from_sequence_pure(model, seq, days)
pred_real = scaler.inverse_transform(pred_scaled)
f_dates = [check_date + timedelta(days=i) for i in range(1, days+1)]
st.markdown("<div class='card'>", unsafe_allow_html=True)
fig = create_combined_chart(df, check_date, f_dates, pred_real)
st.pyplot(fig)
res_df = pd.DataFrame({"Tanggal": [d.strftime("%Y-%m-%d") for d in f_dates], "Harga (USD)": pred_real.flatten()})
st.dataframe(res_df, use_container_width=True)
st.markdown("</div>", unsafe_allow_html=True)
else:
st.error("Tanggal tidak ditemukan dalam dataset.")
st.markdown("</div>", unsafe_allow_html=True)
st.markdown("<div class='footer'>Dibuat oleh Muhammad Gilman Nadhif Azmi</div>", unsafe_allow_html=True)
elif model is None:
st.error("Gagal memuat Model/Scaler. Pastikan file .h5 dan .pkl ada.")