GCP预测结果解读咨询：为何未返回预期的Yes/No输出？

Let me walk you through exactly what each part of this prediction result means, especially since you were expecting a simple Yes/No answer:

{"probabilities": [0.9990405440330505, 0.0009594644070602953], "class_ids": [0], "classes": ["0"], "logits": [-6.948175430297852], "logistic": [0.0009594644070602953]}

Key Field Breakdown: #

• probabilities: This array holds the model's predicted probability for each of your two classes (Yes/No). The first value (0.99904) is the probability for class 0, and the second (0.00096) is for class 1. In short, the model is 99.9% confident in class 0.
• class_ids: This is the model's final predicted category ID. [0] tells you the model picked class 0 as its top choice.
• classes: Just a string-formatted version of class_ids—["0"] matches the ID above, no extra meaning here.
• logits: These are the raw, unnormalized scores from the model's output layer (before applying softmax/sigmoid). The -6.948 here corresponds to class 1, which explains why its probability is so low (lower logits = lower probability in classification tasks).
• logistic: This is the result of applying a sigmoid function to the logit, giving you the probability for one of the two classes. Here, it’s identical to the second value in probabilities, confirming class 1 has a near-zero chance.

Translating to Yes/No: #

You’ll need to cross-reference your training data’s label mapping:

• If class 0 was mapped to No and class 1 to Yes, this prediction translates to a clear No.
• If your mapping was reversed (class 0 = Yes), then it’s Yes—though given the massive probability gap, this is unlikely unless your labeling was swapped during training.

内容的提问来源于stack exchange，提问作者Sofia