基于Softmax的AI模型测试报错：一维数组转二维数组失败

Hey there! Let's walk through the issues in your test code and get your model making predictions properly.

First: The Reshape Error (The Big One) #

Your error happens because you're trying to reshape an 87-element array into a 2-column matrix. 87 is an odd number—you can't split 87 elements evenly into groups of 2!

Your model expects input with shape (number_of_samples, features) (since you set input_dim=features where features=87). For a single sample, that means you need a 2D array of shape (1, 87), not (-1, 2).

Other Key Issues in Your Test Code #

Let's go through the other problems step by step:

• Wrong random data creation: random.rand(features) isn't the right way to create a numpy array of random values—you want np.random.rand(features) instead.
• Mismatched scaler: You used MinMaxScaler during training, but Normalizer in testing. Always use the same scaler, and never fit it on test data (use the scaler you fitted on your training data, or re-load it if you saved it).
• Outdated prediction method: predict_classes() was removed in newer Keras/TensorFlow versions. Use predict() followed by np.argmax() to get class labels.

Fixed Test Code #

Here's the corrected version with explanations:

# Make sure you're using numpy for random data
import numpy as np
from keras.models import Sequential
from keras.layers import Dense
from sklearn import preprocessing

# Recreate the model architecture (same as training)
model2 = Sequential()
model2.add(Dense(400, input_dim=87, activation='tanh'))
model2.add(Dense(150, activation='tanh'))
model2.add(Dense(80, activation='tanh'))
model2.add(Dense(40, activation='tanh'))
model2.add(Dense(5, activation='softmax'))
model2.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# Load the trained weights
model2.load_weights("testSave.hdf5")

# 1. Create a single sample with 87 features (correct numpy syntax)
newData = np.random.rand(87)

# 2. Reshape to (1, 87) to match model input shape (samples, features)
reshaped_data = newData.reshape(1, -1)  # -1 lets numpy calculate the remaining dimension

# 3. Use the SAME scaler as training (MinMaxScaler), don't fit on test data!
# Note: You should save your training scaler during preprocessing, but if not, reinitialize it properly
scaler = preprocessing.MinMaxScaler(feature_range=(0, 1))
# Important: If you have your original training X, use scaler.fit(X) here—don't fit on newData!
# For example: scaler.fit(rescaledX)  # from your training code

# 4. Transform the test data
newDataN = scaler.transform(reshaped_data)

# 5. Make prediction (replace predict_classes with predict + argmax)
predictions = model2.predict(newDataN)
predicted_class = np.argmax(predictions, axis=1)

print("Predicted Class: ", predicted_class[0])

Bonus Tip: Save the Entire Model Instead of Just Weights #

Instead of saving only weights, save the full model architecture + weights together with model.save("my_model.h5"). Then loading is way simpler:

from keras.models import load_model
model2 = load_model("my_model.h5")

This avoids having to redefine the model architecture manually, which reduces mistakes.

内容的提问来源于stack exchange，提问作者RonanD1807