如何在Python的Keras框架中使用训练好的模型确定输入数据？

Hey there! Let's walk through how to use your trained Keras model to make predictions on new input data. First, let's fix a couple of tiny typos in your training code—they'll cause errors if left unaddressed:

Fixed Training Code #

import numpy as np
from keras.models import Sequential
from keras.layers import Dense

np.random.seed(5)
# Fixed: "delimeter" → "delimiter"
dataset = np.loadtxt('path to dataset', delimiter=',')
x_train = dataset[:700, 0:3]
y_train = dataset[:700, 3]
x_test = dataset[700:, 0:3]
y_test = dataset[700:, 3]

model = Sequential()
# Fixed: "activate" → "activation"
model.add(Dense(12, input_dim=3, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
# Completed compile statement with optimizer and metrics
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

# Train the model (you were missing this key step!)
model.fit(x_train, y_train, epochs=100, batch_size=10)

Once your model is trained, you can use it to predict outputs for new input data in three simple steps:

Step 1: Save the Trained Model (Optional but Highly Recommended) #

Saving your model lets you reuse it later without retraining. Add this code right after training finishes:

# Save entire model (structure + weights + training config) in one file
model.save('my_trained_model.h5')

# Alternative: Save structure and weights separately (if you prefer)
# model_json = model.to_json()
# with open("model_structure.json", "w") as json_file:
#     json_file.write(model_json)
# model.save_weights("model_weights.h5")

Step 2: Load the Saved Model (If You Saved It) #

If you closed your Python session and want to reuse the model later, load it like this:

from keras.models import load_model

# Load the full saved model
loaded_model = load_model('my_trained_model.h5')

# If you saved structure/weights separately:
# from keras.models import model_from_json
# json_file = open('model_structure.json', 'r')
# loaded_model_json = json_file.read()
# json_file.close()
# loaded_model = model_from_json(loaded_model_json)
# loaded_model.load_weights("model_weights.h5")
# # Re-compile the model (required for separate saves)
# loaded_model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

Step 3: Make Predictions on New Input Data #

Your model expects input data with the same shape as your training data (3 features per sample, formatted as a 2D array). Here's how to use it:

Example 1: Single Input Sample #

# Create a 2D array (even for one sample—Keras expects batch inputs)
new_sample = np.array([[1.5, 2.3, 4.1]])

# Get prediction probability (from sigmoid activation)
pred_prob = loaded_model.predict(new_sample)
# Convert probability to binary class (0 or 1) using 0.5 as threshold
pred_class = (pred_prob > 0.5).astype(int)

print(f"Predicted Probability: {pred_prob[0][0]:.4f}")
print(f"Predicted Class: {pred_class[0][0]}")

Example 2: Multiple Input Samples #

# Batch of 3 samples, each with 3 features
batch_samples = np.array([
    [0.2, 0.4, 0.6],
    [7.2, 8.1, 9.3],
    [3.5, 1.7, 2.9]
])

pred_probs = loaded_model.predict(batch_samples)
pred_classes = (pred_probs > 0.5).astype(int)

for idx, (prob, cls) in enumerate(zip(pred_probs, pred_classes)):
    print(f"Sample {idx+1}: Probability = {prob[0]:.4f}, Class = {cls[0]}")

Key Notes #

• Match Preprocessing: If you normalized/standardized your training data (e.g., scaled features to 0-1), apply the exact same transformation to new input data—otherwise predictions will be unreliable.
• Input Shape: Always pass input as a 2D array. A shape like (3,) will cause errors; use (1, 3) for single samples instead.
• No Save? No Problem: If you're using the model immediately after training, you can skip saving/loading and just call model.predict() directly with your new data.

内容的提问来源于stack exchange，提问作者송준석