Datasets:
File size: 18,510 Bytes
4378303 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 |
import requests
import pandas as pd
import mplfinance as mpf
import matplotlib.pyplot as plt
from datetime import datetime, timedelta
import os
import numpy as np
from PIL import Image
import tensorflow as tf
from tensorflow.keras import layers, models
from sklearn.metrics import accuracy_score, f1_score, recall_score, roc_auc_score, precision_recall_curve, auc
from sklearn.utils.class_weight import compute_class_weight
import argparse
import gc
import time
# Use non-interactive backend for matplotlib
plt.switch_backend('Agg')
# Coin configurations
COINS = {
"BTCUSDT": {"train_month": (2024, 6), "test_months": [(2024, 12), (2024, 3), (2024, 8), (2024, 4), (2024, 1)]},
"ETHUSDT": {"train_month": (2024, 6), "test_months": [(2024, 8), (2024, 4), (2024, 5), (2024, 3), (2024, 2)]},
"BNBUSDT": {"train_month": (2024, 10), "test_months": [(2024, 3), (2024, 12), (2024, 8), (2024, 1), (2024, 4)]},
"XRPUSDT": {"train_month": (2024, 9), "test_months": [(2024, 11), (2024, 12), (2024, 4), (2024, 8), (2024, 1)]},
"ADAUSDT": {"train_month": (2024, 9), "test_months": [(2024, 4), (2024, 12), (2024, 1), (2024, 3), (2024, 11)]},
"DOGEUSDT": {"train_month": (2024, 9), "test_months": [(2024, 3), (2024, 4), (2024, 11), (2024, 8), (2024, 12)]}
}
TIME_LENGTHS = [7, 14, 21, 28] # 1, 2, 3, 4 weeks in days
WINDOW_SIZES = [5, 15, 30] # Candles per image
MISSING_RATIOS = [0.6, 0.8, 0.95] # 60%, 80%, 95% missing data
# Set BASE_DIR to new folder for irregular data
BASE_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "crypto_research_minute_irregular")
# Binance API data fetcher with irregular data omission
def fetch_coin_data(symbol, start_time, end_time, missing_ratio):
url = "https://api.binance.com/api/v3/klines"
all_data = []
current_start = int(start_time.timestamp() * 1000)
end_ms = int(end_time.timestamp() * 1000)
while current_start < end_ms:
params = {"symbol": symbol, "interval": "1m", "startTime": current_start, "endTime": end_ms, "limit": 1000}
response = requests.get(url, params=params)
data = response.json()
if not data:
break
all_data.extend(data)
current_start = int(data[-1][0]) + 60000 # 1 minute in milliseconds
df = pd.DataFrame(all_data, columns=["timestamp", "open", "high", "low", "close", "volume", "close_time", "quote_asset_volume", "trades", "taker_buy_base", "taker_buy_quote", "ignore"])
df["timestamp"] = pd.to_datetime(df["timestamp"], unit="ms")
df[["open", "high", "low", "close"]] = df[["open", "high", "low", "close"]].astype(float)
# Apply irregular data omission
if missing_ratio > 0:
n_rows = len(df)
n_keep = int(n_rows * (1 - missing_ratio))
if n_keep < 1: # Allow at least 1 row
print(f"Warning: Not enough data after {missing_ratio*100}% omission for {symbol}, keeping all data")
return df[["timestamp", "open", "high", "low", "close"]]
keep_indices = np.random.choice(n_rows, size=n_keep, replace=False)
df = df.iloc[keep_indices].sort_values("timestamp").reset_index(drop=True)
return df[["timestamp", "open", "high", "low", "close"]]
# Generate candlestick images and labels with sparse windows
def generate_images(df, window_size, output_dir, period_name, month_str, missing_ratio):
os.makedirs(output_dir, exist_ok=True)
labels_file = os.path.join(output_dir, f"labels_{month_str}_1m_{period_name}_w{window_size}_{int(missing_ratio*100)}pct.csv")
if os.path.exists(labels_file):
print(f"Labels already exist at {labels_file}, skipping image generation")
return labels_file
if len(df) < 1:
print(f"Warning: DataFrame too small ({len(df)} rows) for any window, skipping image generation")
return None
labels = []
start_time = time.time()
# Use index as timestamps since it's set as index
original_timestamps = pd.date_range(start=df.index[0], end=df.index[-1], freq="1min")
for i in range(len(original_timestamps) - window_size + 1):
window_start = original_timestamps[i]
window_end = original_timestamps[i + window_size - 1]
window_indices = df.index[(df.index >= window_start) & (df.index <= window_end)]
window_df = df.loc[window_indices]
if len(window_df) > 0:
first_candle = window_df.iloc[0]
last_candle = window_df.iloc[-1]
label = "UP" if last_candle["close"] > first_candle["open"] else "DOWN"
labels.append(label)
plt.figure(figsize=(2, 2))
mpf.plot(window_df, type="candle", style="binance", axisoff=True, title="", ylabel="", xlabel="", volume=False, tight_layout=True)
image_path = os.path.join(output_dir, f"candle_{i}_{int(missing_ratio*100)}pct.png")
plt.savefig(image_path, bbox_inches="tight", pad_inches=0, dpi=32)
plt.close('all')
if i % 1000 == 0:
elapsed = time.time() - start_time
images_generated = i + 1
speed = images_generated / elapsed if elapsed > 0 else 0
print(f"Generated image {i}/{len(original_timestamps) - window_size + 1} for {month_str} 1m {period_name} w{window_size} {missing_ratio*100}% ({speed:.2f} images/sec)")
else:
continue
labels_df = pd.DataFrame({"image_path": [f"candle_{i}_{int(missing_ratio*100)}pct.png" for i in range(len(original_timestamps) - window_size + 1) if os.path.exists(os.path.join(output_dir, f"candle_{i}_{int(missing_ratio*100)}pct.png"))], "label": labels})
labels_df.to_csv(labels_file, index=False)
print(f"Saved {len(labels_df)} labels to {labels_file}")
return labels_file
# Load and preprocess images
def load_images(labels_file, images_dir):
if not os.path.exists(labels_file):
return None, None
labels_df = pd.read_csv(labels_file)
X = np.array([np.array(Image.open(os.path.join(images_dir, row["image_path"])).convert("RGB").resize((64, 64))) / 255.0 for _, row in labels_df.iterrows()])
y = np.array([1 if label == "UP" else 0 for label in labels_df["label"]])
return X, y
# Train CNN model
def train_model(X, y, period_name, month_str, window_size, coin_dir, missing_ratio):
model_path = os.path.join(coin_dir, "models", f"model_{month_str}_1m_{period_name}_w{window_size}_{int(missing_ratio*100)}pct.h5")
if os.path.exists(model_path):
print(f"Model already exists at {model_path}, loading instead of training")
return tf.keras.models.load_model(model_path), None
model = models.Sequential([
layers.Conv2D(32, (3, 3), activation="relu", input_shape=(64, 64, 3)),
layers.MaxPooling2D((2, 2)),
layers.Dropout(0.25),
layers.Conv2D(64, (3, 3), activation="relu"),
layers.MaxPooling2D((2, 2)),
layers.Dropout(0.25),
layers.Conv2D(128, (3, 3), activation="relu"),
layers.Flatten(),
layers.Dense(128, activation="relu"),
layers.Dropout(0.5),
layers.Dense(1, activation="sigmoid")
])
model.compile(optimizer="adam", loss="binary_crossentropy", metrics=["accuracy"])
class_weights = compute_class_weight('balanced', classes=np.unique(y), y=y)
history = model.fit(X, y, epochs=10, batch_size=32, class_weight=dict(enumerate(class_weights)))
model.save(model_path)
print(f"Model saved to {model_path}")
return model, history
# Evaluate and save results
def evaluate_and_save(model, X, y, period_name, month_str, window_size, coin_dir, dataset_type="train", exp_suffix="", missing_ratio=0):
results_file = os.path.join(coin_dir, "results", f"results_{dataset_type}_{month_str}_1m_{period_name}_w{window_size}_{int(missing_ratio*100)}pct{exp_suffix}.txt")
if os.path.exists(results_file) and exp_suffix != "_exp2":
print(f"Results already exist at {results_file}, skipping evaluation")
return None
y_pred_prob = model.predict(X, verbose=0)
y_pred = (y_pred_prob > 0.5).astype(int).flatten()
metrics = {
"accuracy": accuracy_score(y, y_pred),
"f1": f1_score(y, y_pred),
"recall": recall_score(y, y_pred),
"auroc": roc_auc_score(y, y_pred_prob),
"auprc": auc(*precision_recall_curve(y, y_pred_prob)[1::-1])
}
with open(results_file, "w") as f:
f.write(f"{dataset_type.capitalize()} Metrics for {month_str} 1m {period_name} w{window_size} {missing_ratio*100}% {exp_suffix}:\n")
for k, v in metrics.items():
f.write(f"{k.capitalize()}: {v:.4f}\n")
print(f"Results saved to {results_file}")
return metrics
# Check if all experiments for a window size and missing ratio are complete
def is_window_size_complete(symbol, train_month, test_months, window_size, missing_ratio):
coin_dir = os.path.join(BASE_DIR, symbol)
train_year, train_month_num = train_month
train_month_str = f"{train_year}-{train_month_num:02d}"
ratio_str = f"_{int(missing_ratio*100)}pct"
# Check Experiment I
for days in TIME_LENGTHS:
period_name = f"{days}days"
train_result = os.path.join(coin_dir, "results", f"results_train_{train_month_str}_1m_{period_name}_w{window_size}{ratio_str}.txt")
if not os.path.exists(train_result):
return False
for test_year, test_month_num in test_months:
test_month_str = f"{test_year}-{test_month_num:02d}"
test_result = os.path.join(coin_dir, "results", f"results_test_{test_month_str}_1m_{period_name}_w{window_size}{ratio_str}.txt")
if not os.path.exists(test_result):
return False
# Check Experiment II
period_name = "1week"
train_result = os.path.join(coin_dir, "results", f"results_train_{train_month_str}_1m_{period_name}_w{window_size}{ratio_str}_exp2.txt")
if not os.path.exists(train_result):
return False
for test_year, test_month_num in test_months:
test_month_str = f"{test_year}-{test_month_num:02d}"
for days in [14, 21, 28]:
period_name = f"{days}days"
test_result = os.path.join(coin_dir, "results", f"results_test_{test_month_str}_1m_{period_name}_w{window_size}{ratio_str}_exp2.txt")
if not os.path.exists(test_result):
return False
return True
# Main experiment runner for a single coin, window size, and missing ratio
def run_experiments_for_coin(symbol, train_month, test_months, window_size, missing_ratio):
if is_window_size_complete(symbol, train_month, test_months, window_size, missing_ratio):
print(f"All experiments for {symbol} with window size {window_size} and {missing_ratio*100}% missing complete, skipping")
return
coin_dir = os.path.join(BASE_DIR, symbol)
RAW_DATA_DIR = os.path.join(coin_dir, "raw_data")
IMAGES_DIR = os.path.join(coin_dir, "images")
MODELS_DIR = os.path.join(coin_dir, "models")
RESULTS_DIR = os.path.join(coin_dir, "results")
os.makedirs(RAW_DATA_DIR, exist_ok=True)
os.makedirs(IMAGES_DIR, exist_ok=True)
os.makedirs(MODELS_DIR, exist_ok=True)
os.makedirs(RESULTS_DIR, exist_ok=True)
train_year, train_month_num = train_month
ratio_str = f"_{int(missing_ratio*100)}pct"
# Experiment I: Train and test on matching timelengths
for days in TIME_LENGTHS:
period_name = f"{days}days"
train_start = datetime(train_year, train_month_num, 1)
train_end = train_start + timedelta(days=days - 1, hours=23, minutes=59)
train_month_str = f"{train_year}-{train_month_num:02d}"
raw_file = os.path.join(RAW_DATA_DIR, f"raw_{train_month_str}_1m_{period_name}{ratio_str}.csv")
if not os.path.exists(raw_file):
df = fetch_coin_data(symbol, train_start, train_end, missing_ratio)
df.set_index("timestamp", inplace=True)
df.to_csv(raw_file)
print(f"Raw data saved to {raw_file}")
else:
print(f"Raw data already exists at {raw_file}, skipping fetch")
df = pd.read_csv(raw_file, index_col="timestamp", parse_dates=["timestamp"])
df.index = pd.to_datetime(df.index)
images_subdir = os.path.join(IMAGES_DIR, f"{train_month_str}_1m_{period_name}_w{window_size}{ratio_str}")
labels_file = generate_images(df, window_size, images_subdir, period_name, train_month_str, missing_ratio)
if labels_file:
X, y = load_images(labels_file, images_subdir)
if X is not None:
model, history = train_model(X, y, period_name, train_month_str, window_size, coin_dir, missing_ratio)
evaluate_and_save(model, X, y, period_name, train_month_str, window_size, coin_dir, "train", missing_ratio=missing_ratio)
tf.keras.backend.clear_session()
gc.collect()
for test_year, test_month_num in test_months:
test_start = datetime(test_year, test_month_num, 1)
test_end = test_start + timedelta(days=days - 1, hours=23, minutes=59)
test_month_str = f"{test_year}-{test_month_num:02d}"
raw_file = os.path.join(RAW_DATA_DIR, f"raw_{test_month_str}_1m_{period_name}{ratio_str}.csv")
if not os.path.exists(raw_file):
df = fetch_coin_data(symbol, test_start, test_end, missing_ratio)
df.set_index("timestamp", inplace=True)
df.to_csv(raw_file)
print(f"Raw data saved to {raw_file}")
else:
print(f"Raw data already exists at {raw_file}, skipping fetch")
df = pd.read_csv(raw_file, index_col="timestamp", parse_dates=["timestamp"])
df.index = pd.to_datetime(df.index)
images_subdir = os.path.join(IMAGES_DIR, f"{test_month_str}_1m_{period_name}_w{window_size}{ratio_str}")
labels_file = generate_images(df, window_size, images_subdir, period_name, test_month_str, missing_ratio)
if labels_file:
X, y = load_images(labels_file, images_subdir)
if X is not None:
evaluate_and_save(model, X, y, period_name, test_month_str, window_size, coin_dir, "test", missing_ratio=missing_ratio)
tf.keras.backend.clear_session()
gc.collect()
# Experiment II: Train on 1 week, test on 2-3-4 weeks
exp2_test_lengths = [14, 21, 28]
train_start = datetime(train_year, train_month_num, 1)
train_end = train_start + timedelta(days=6, hours=23, minutes=59)
train_month_str = f"{train_year}-{train_month_num:02d}"
period_name = "1week"
raw_file = os.path.join(RAW_DATA_DIR, f"raw_{train_month_str}_1m_{period_name}{ratio_str}.csv")
if not os.path.exists(raw_file):
df = fetch_coin_data(symbol, train_start, train_end, missing_ratio)
df.set_index("timestamp", inplace=True)
df.to_csv(raw_file)
print(f"Raw data saved to {raw_file}")
else:
print(f"Raw data already exists at {raw_file}, skipping fetch")
df = pd.read_csv(raw_file, index_col="timestamp", parse_dates=["timestamp"])
df.index = pd.to_datetime(df.index)
images_subdir = os.path.join(IMAGES_DIR, f"{train_month_str}_1m_{period_name}_w{window_size}{ratio_str}")
labels_file = generate_images(df, window_size, images_subdir, period_name, train_month_str, missing_ratio)
if labels_file:
X, y = load_images(labels_file, images_subdir)
if X is not None:
model, history = train_model(X, y, period_name, train_month_str, window_size, coin_dir, missing_ratio)
evaluate_and_save(model, X, y, period_name, train_month_str, window_size, coin_dir, "train", "_exp2", missing_ratio)
tf.keras.backend.clear_session()
gc.collect()
for test_year, test_month_num in test_months:
test_month_str = f"{test_year}-{test_month_num:02d}"
for days in exp2_test_lengths:
period_name = f"{days}days"
test_start = datetime(test_year, test_month_num, 1)
test_end = test_start + timedelta(days=days - 1, hours=23, minutes=59)
raw_file = os.path.join(RAW_DATA_DIR, f"raw_{test_month_str}_1m_{period_name}{ratio_str}.csv")
if not os.path.exists(raw_file):
df = fetch_coin_data(symbol, test_start, end_time, missing_ratio)
df.set_index("timestamp", inplace=True)
df.to_csv(raw_file)
print(f"Raw data saved to {raw_file}")
else:
print(f"Raw data already exists at {raw_file}, skipping fetch")
df = pd.read_csv(raw_file, index_col="timestamp", parse_dates=["timestamp"])
df.index = pd.to_datetime(df.index)
images_subdir = os.path.join(IMAGES_DIR, f"{test_month_str}_1m_{period_name}_w{window_size}{ratio_str}")
labels_file = generate_images(df, window_size, images_subdir, period_name, test_month_str, missing_ratio)
if labels_file:
X, y = load_images(labels_file, images_subdir)
if X is not None:
evaluate_and_save(model, X, y, period_name, test_month_str, window_size, coin_dir, "test", "_exp2", missing_ratio)
tf.keras.backend.clear_session()
gc.collect()
# Run experiments for all coins, window sizes, and missing ratios
def run_all_experiments():
os.makedirs(BASE_DIR, exist_ok=True)
for symbol, config in COINS.items():
for window_size in WINDOW_SIZES:
for missing_ratio in MISSING_RATIOS:
print(f"Running experiments for {symbol} with window size {window_size} and {missing_ratio*100}% missing")
run_experiments_for_coin(symbol, config["train_month"], config["test_months"], window_size, missing_ratio)
print(f"Completed experiments for {symbol} with window size {window_size} and {missing_ratio*100}% missing")
tf.keras.backend.clear_session()
gc.collect()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Crypto Minute-Based Image Classification with Irregular Missing Data and Sparse Windows")
args = parser.parse_args()
run_all_experiments() |