python中的MAPE（平均绝对百分比误差）测量导致错误

Question

我正在尝试在我的随机森林代码中测量 MAPE（平均绝对百分比误差）值。 MAE 值为 7.5。当我尝试计算 MAPE 时，它会输出：

Accuracy: -inf %

这是我计算 MAPE 的代码。如何使其工作或为什么不计算值。

mape = 100 * (errors / test_labels)
# Calculate and display accuracy
accuracy = 100 - np.mean(mape)
print('Accuracy:', round(accuracy, 2), '%.')

以下是值：

 errors: array([ 2.165,  6.398,  2.814, ..., 21.268,  8.746, 11.63 ])
 test_labels: array([45, 47, 98, ..., 87, 47, 72])

这些是类型：

var1          int64
var2          int64
var3          float64
var4          int64
var6          float64
var7          int64
var1.         float64
dtype: object

示例值，超过 8000 个条目

      var1  var2.   var3               var4     var5                var6    var7
"420823370" "183"   "2019-09-07 22:13:04"   "84"    "2019-09-07 22:12:46"   "72"    "00:00:18"
"420521201" "183"   "2019-09-07 17:43:03"   "84"    "2019-09-07 17:42:51"   "46"    "00:00:12"
"420219554" "183"   "2019-09-07 12:43:02"   "88"    "2019-09-07 12:42:39"   "72"    "00:00:23"
"419618820" "183"   "2019-09-07 02:43:01"   "92"    "2019-09-07 02:42:46"   "80"    "00:00:15"
"419618819" "183"   "2019-09-07 02:43:01"   "84"    "2019-09-07 02:42:46"   "80"    "00:00:15"
"417193989" "183"   "2019-09-05 10:42:52"   "82"    "2019-09-05 10:42:23"   "0" "00:00:29"
"416891691" "183"   "2019-09-05 05:42:51"   "78"    "2019-09-05 05:42:49"   "72"    "00:00:02"
"416587222" "183"   "2019-09-05 00:42:51"   "88"    "2019-09-05 00:42:35"   "99"    "00:00:16"
"416587223" "183"   "2019-09-05 00:42:51"   "82"    "2019-09-05 00:42:35"   "99"    "00:00:16"
"416587224" "183"   "2019-09-05 00:42:51"   "80"    "2019-09-05 00:42:35"   "99"    "00:00:16"

id:Big Int. ts_tuid: Big Int. rssi: numeric. batl: real. ts_diff:interval 

这里是代码示例：

从 CSV 加载数据

model = (
    pd.read_csv("source.csv", parse_dates=['var3', 'var5'], date_parser=lambda x: pd.to_datetime(x))
    .assign(
        rssi_ts=lambda x: x.loc[:, 'var3'].astype(int) / 10 ** 9,
        batl_ts=lambda x: x.loc[:, 'var5'].astype(int) / 10 ** 9,
        ts_diff=lambda x: pd.to_timedelta(x.loc[:, 'ts_diff']).astype(int) / 10 ** 9
    )
)

# Labels are the values we want to predict
labels_b = np.array(halti['var4'])
# Remove the labels from the features
# axis 1 refers to the columns
features_r = halti.drop('var4', axis = 1)
features_r2 = list(features_r.columns) 
# Convert to numpy array
features_r = np.array(features_r)

# Using Skicit-learn to split data into training and testing sets
from sklearn.model_selection import train_test_split
# Split the data into training and testing sets
train_features, test_features, train_labels, test_labels = train_test_split(features_r, labels_b, test_size = 0.25, random_state = 42)

# Import the model we are using
from sklearn.ensemble import RandomForestRegressor
# Instantiate model with 1000 decision trees
rf = RandomForestRegressor(n_estimators = 1000, random_state = 42)
# Train the model on training data
rf.fit(train_features, train_labels);

# Use the forest's predict method on the test data
predictions = rf.predict(test_features)
# Calculate the absolute errors
errors = abs(predictions - test_labels)
# Print out the mean absolute error (mae)
print('Mean Absolute Error:', round(np.mean(errors), 2), 'degrees.')

mape = 100 * (errors / test_labels)
# Calculate and display accuracy
accuracy = 100 - np.mean(mape)
print('Accuracy:', round(accuracy, 2), '%.')

Answer 1

这次输出显示 Inf 在 mape 错误度量中。其背后的原因是我们的观测值为零。当因变量可以将零作为输出之一时，我们不能使用 mape 作为误差度量。在这种情况下，应该使用其他错误度量。

参考：https://rstudio-pubs-static.s3.amazonaws.com/390751_f6b763e827b24c9cb4406cd43129c8a9.html

Answer 2

您收到此错误是因为当您的测试标签为 0 时 MAPE 未定义，这是几个 shortcomings of using MAPE 之一。如果您将accuracy = 100 - np.mean(mape) 替换为accuracy = 100 - np.mean(mape[np.isfinite(mape)])，您将得到一个更合理的数字。