def convert_pm(pm, us_pm, r):
crawling = Crawling()
pm_ = []
# extract stations class
for d in pm:
pr = []
for s in us_pm:
v = d[s] * r
#pr.append(get_aqi_class(crawling.AQIPM25(v)))
pr.append(crawling.aqi_pm25_china_class(v))
pm_.append(pr)
# extract class for all pixels
# for d in pm:
# pr = []
# for idx in d:
# pr.append(get_aqi_class(crawling.AQIPM25(idx)))
# pm_.append(pr)
return pm_
def evaluate_china(url,
url2,
stations,
encoder_length=12,
decoder_length=24,
eval_stat=False):
cr = Crawling()
# missing = [136,156,824,1028,1053,1084,1085,1457,1460,1461,1462,1464,1465,1476,1477, \
# 1478,1479,1481,1482,1484,1485,1486,1487,1510,1841,3814,4246,4268,4301,4311, \
# 4313,4317,4329,4331,4333,4349,4360,4504,4524,4529,4532,4535,4580,4860,5270, \
# 5457,5489,5509,5962,6007,6011,6039,6054,6125,6172,6189,6192,6201,6230,6234, \
# 6246,6254,6255,6256,6257,6258,6295,6300,6319,6336,6356,6362,6371,6372,6394, \
# 6491,6492,6493,6494,6497,6517,6519,6523,6539,6559,6564,6568,6569,6570,6612, \
# 6613,6614,6615,6616,6617,6618,6619,6620,6621,6622,6623,6624,6625,6626,6627, \
# 6628,6629,6630,6631,6632,6633,6634,6635,6636,6637,6638,6687,6704,6754,6832, \
# 6849,6858,6873,6979,7255,8099,8100,8101,8225,8226,8227,8228,8229,8230,8231, \
# 8232,8233,8234,8235,8236,8237,8238,8239,8240,8241,8273,8274,8275,8276,8277, \
# 8278,8279,8280,8281,8282,8283,8284,8285,8286,8287,8288,8289,8290,8291,8344, \
# 8421,8422,8423,8424,8425,8426,8427,8428,8429,8430,8431,8432,8495,8496,8497, \
# 8498,8499,8500,8501,8502,8503,8504,8631,8759]
data = utils.load_file(url)
labels = utils.load_file(url2)
if type(data) is list:
data = np.asarray(data)
if len(data.shape) == 4:
lt = data.shape[0] * data.shape[1]
data = np.reshape(data, (lt, data.shape[-2], data.shape[-1]))
labels = np.reshape(labels, (lt, data.shape[-2], data.shape[-1]))
else:
lt = len(data)
data = np.array(data)
labels = np.array(labels)
if not eval_stat:
loss_mae = [0.0] * decoder_length
else:
loss_mae = np.array([[0.0] * decoder_length] * len(stations))
classes = [0.0] * decoder_length
# loss_lt = 0
loss_total = [0.0] * decoder_length
# ckt = 0
valid = []
for i, (d_, l_) in enumerate(zip(data, labels)):
# st = 2 * i
# ed = st + decoder_length
# flag = False
# for ck in range(st, ed):
# if ck in missing:
# flag = True
# break
# if flag:
# continue
d = d_[:decoder_length, :]
lbt = l_[:decoder_length, :]
for t_i, (t, l_t) in enumerate(zip(d, lbt)):
pred_t = []
label_t = []
for r in stations:
#pred_t.append(t[r])
#label_t.append(l_t[r])
# if not math.isnan(t[r]):
pr = t[r]
if pr < 0:
pr = 0
print(pr)
pred_t.append(cr.aqi_pm25_china(pr))
label_t.append(cr.aqi_pm25_china(l_t[r] * 300))
# label_t.append(l_t[r])
# print(t[r], l_t[r])
# if pred_t:
if not eval_stat:
mae = mean_absolute_error(pred_t, label_t)
classes[t_i] += accuracy_score(
[cr.aqi_pm25_china_class(x) for x in pred_t],
[cr.aqi_pm25_china_class(x) for x in label_t])
if mae > 80:
valid.append("%i,%i" % (i, t_i))
if mae < 80:
loss_total[t_i] += 1
loss_mae[t_i] += mae
else:
mae = [abs(p_s - l_s) for p_s, l_s in zip(pred_t, label_t)]
for m_i, m in enumerate(mae):
loss_mae[m_i, t_i] += m
# loss_lt += 1
# utils.update_progress((i + 1.0) / lt)
va = "\n".join(valid)
save_file("outline_china", va, False)
loss_mae = [x / y if y > 0 else 0 for x, y in zip(loss_mae, loss_total)]
for i, x in enumerate(loss_mae):
# if not eval_stat:
print(x, loss_total[i])
# else:
# print([y/loss_lt for y in x])
# print("accumulated:", np.mean(loss_mae[1:6]))
print(np.array(classes) / lt)