Keras：预测不准确

Question

我有一个回归问题，我试图从Excel工作表中预测列set_random_seed(7) df = pd.read_excel("ABC_sheet.xlsx") data = df.iloc[:, 0:2] labels_column = df['C'] training_data = data training_labels = labels_column print("Creating the pickle_model ...") model = Sequential() model.add(Dense(300, input_dim=2, activation='relu')) model.add(Dense(300, activation='relu')) model.add(Dense(300, activation='relu')) model.add(Dense(300, activation='relu')) model.add(Dense(1, activation='sigmoid')) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) model.summary() print("pickle_models Fitting ...") model.fit(training_data, training_labels, epochs=20000, verbose=1) print("Training has been completed successfully!") # Expected Res: 0.00531548914829121 Xnew0 = np.array([[0.54, 0.63446746]]) # Expected Res: 6.34934075187519E-07 Xnew = np.array([[0.44, 0.37447464]]) # Expected Res: 0.0000628 Xnew2 = np.array([[0.53, 0.00063189]]) # Expected Res: 0.00531548914829121 Xnew3 = np.array([[0.54, 0.63446746]]) ynew = model.predict(Xnew0) ynew2 = model.predict(Xnew) ynew3 = model.predict(Xnew2) ynew5 = model.predict(Xnew3) print("Predicted1 =%s, Predicted2 =%s, Predicted3 =%s, Predicted4 =%s, Predicted4 =%s" % (ynew[0], ynew2[0], ynew3[0], ynew5[0])) 的值（附有图片）。

Predicted1 =[0.00591827], Predicted2 =[6.762405e-08], Predicted3 =[0.00139873], Predicted4 =[0.00591827]

输出：

Predicted3 =[0.00139873]

结果与预期结果不符。由于Expected Res: 0.0000628应该更接近def GetNormalizedValue(val, min, max): if val > max: val = max if val < min: val = min if min == max: return 0 denominator = max - min numerator = float(val) - min value = numerator / denominator # print(value) return value def NormalizeData(df): for index, row in df.iterrows(): col_A = row['A'] col_B = row['B'] col_C = row['C'] column_A = GetNormalizedValue(col_A, 0, 100) column_B = GetNormalizedValue(col_B, 0, 1000000) column_C = GetNormalizedValue(col_C, 1, 10000)

我也尝试使用不同数量的层/神经元。有什么帮助吗？

已编辑：

System.setProperty("webdriver.chrome.verboseLogging", "false");