test_shuffled_target_decoding.py

import numpy as np
import pandas as pd

# Load data from the shuffled .npy files
X_path = "datasets/hdc_pairs_200_shot/X_pairs_shuffled.npy"
Y_path = "datasets/hdc_pairs_200_shot/Y_pairs_shuffled.npy"
O_path = "datasets/hdc_pairs_200_shot/O_pairs_shuffled.npy"

X = np.load(X_path, mmap_mode="r")
Y = np.load(Y_path, mmap_mode="r")
O = np.load(O_path, mmap_mode="r")

# Choose a random index
idx = np.random.randint(0, len(Y))
print(f"\n🔀 Random sample index selected: {idx}")

# Get the sample
y_sample = Y[idx][:10]
o_sample = O[idx]

# Decode using same logic as inference
VOLUME_K = 0.2
OCLOSE   = 0
OVOLUME  = 1
OQUOTE   = 2
ONTRADES = 3
OTBBAV   = 4
OTBQAV   = 5

prev_close = float(o_sample[OCLOSE])
prev_vals = {
    "volume": float(o_sample[OVOLUME]),
    "quote_asset_volume": float(o_sample[OQUOTE]),
    "number_of_trades": float(o_sample[ONTRADES]),
    "taker_buy_base_asset_volume": float(o_sample[OTBBAV]),
    "taker_buy_quote_asset_volume": float(o_sample[OTBQAV]),
}

decoded_rows = []
for row in y_sample:
    # OHLC decoding
    dec = {
        "open":  prev_close * np.exp(row[0]),
        "high":  prev_close * np.exp(row[1]),
        "low":   prev_close * np.exp(row[2]),
        "close": prev_close * np.exp(row[3]),
    }
    # Volume decoding
    vol_keys = [
        "volume", "quote_asset_volume", "number_of_trades",
        "taker_buy_base_asset_volume", "taker_buy_quote_asset_volume"
    ]
    for i, key in enumerate(vol_keys, start=4):
        log_ret = np.arctanh(np.clip(row[i], -0.999999, 0.999999)) / VOLUME_K
        dec[key] = prev_vals[key] * np.exp(log_ret)

    # Update previous values for next step
    prev_close = dec["close"]
    for key in prev_vals:
        prev_vals[key] = dec[key]

    decoded_rows.append(dec)

# Display in terminal
decoded_df = pd.DataFrame(decoded_rows)
print("\n📋 Decoded first 10 target rows:")
print(decoded_df.to_string(index=False))