"""
Generator to read in scraped profiles and return batches.
Returns:
 - X:   Curve data (BATCH_SIZE, POINTS, 2, 1) in cartesian coordinates
 - y:   Parameters (3, 1) (normalized to mean and stddev (cd was scaled by log to get normal distribution))
"""
"""

生成器读取报废的配置文件并返回批次。

退货：

-X：笛卡尔坐标中的曲线数据（BATCH_SIZE，POINTS，2，1）

-y：参数（3，1）（归一化为平均值和标准偏差（cd按对数缩放以获得正态分布））

"""

################################################################################
# %% 引入包
################################################################################

import glob
import numpy as np

################################################################################
# %% 定义发电机功能
################################################################################

def profile_generator(BATCH_SIZE=512, POINTS=64, DTYPE='float32'):

    fimg = sorted(glob.glob(f'01-prep-data/{POINTS}/X_*.npy'))

    #print(f'----> Reading {fimg[0]} <----')
    X = np.load(fimg[0])
    y = np.load(fimg[0].replace('X', 'y'))

    idx = np.random.permutation(len(X))
    X = X[idx]
    y = y[idx]

    i = 0

    while True:

        if len(X) < BATCH_SIZE:

            if i<len(fimg)-1:
                i += 1
            else:
                i = 0

            #print(f'----> Reading {fimg[i]} <----')
            X = np.concatenate((X, np.load(fimg[i])), axis=0)
            y = np.concatenate((y, np.load(fimg[i].replace('X', 'y'))), axis=0)

            idx = np.random.permutation(len(X))
            X = X[idx]
            y = y[idx]

        else:

            ##### SHIFT/SCALE TO -1.0 ... 1.0
            X[:BATCH_SIZE,: ,0, 0] = (X[:BATCH_SIZE,:, 0, 0] - 0.5)*2.0
            X[:BATCH_SIZE,: ,1, 0] = (X[:BATCH_SIZE,:, 1, 0])*2.0

            ##### CL
            ymean = 0.5
            ystd =  0.7
            y[:BATCH_SIZE, 0] = (y[:BATCH_SIZE, 0]-ymean)/ystd

            ##### CD
            ymean = -3.6
            ystd = 0.7
            y[:BATCH_SIZE, 1] = (np.log(y[:BATCH_SIZE, 1])-ymean)/ystd

            ymean = 0.085
            ystd = 0.025
            y[:BATCH_SIZE, 2] = (y[:BATCH_SIZE, 2]-ymean)/ystd

            yield X[:BATCH_SIZE].astype(DTYPE), y[:BATCH_SIZE].astype(DTYPE)

            X = np.delete(X, range(BATCH_SIZE), axis=0)
            y = np.delete(y, range(BATCH_SIZE), axis=0)