Hey there! Let's work through this frustrating error you're hitting. That TypeError pops up when you try to use a numpy array in a spot that expects a hashable object—think dictionary keys, set elements, or even some TensorFlow API calls that rely on hashable values under the hood. Let's break down the most common causes and fixes:

Common Scenarios & Fixes #

1. You're using a numpy array as a dictionary key or set element #

This is the most frequent culprit. Numpy arrays are mutable, so they can't be hashed (a requirement for dictionary keys and set members).

Example of the error:

import numpy as np
my_array = np.array([1, 2, 3])
my_dict = {}
my_dict[my_array] = "model_weights"  # Boom! Error hits here

Fix it by converting the array to a tuple:

Tuples are immutable and hashable, so they work perfectly for this use case:

my_dict[tuple(my_array)] = "model_weights"

2. Incorrect input handling when passing data to TensorFlow #

If the error fires while training or running your model, you're probably passing numpy arrays in a wrong format to TensorFlow APIs. For example:

• Accidentally using a numpy array as a key in a dictionary when calling model.fit()
• Passing a nested array structure that TensorFlow can't process without hash checks

Example of the error:

import tensorflow as tf
import numpy as np

X_train = np.random.rand(100, 10)
y_train = np.random.randint(0, 2, size=(100,))

model = tf.keras.Sequential([
    tf.keras.layers.Dense(32, activation='relu', input_shape=(10,)),
    tf.keras.layers.Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam', loss='binary_crossentropy')
model.fit({X_train: y_train}, epochs=5)  # Wrong! Using X_train as a dict key

Fix it by passing inputs and labels correctly:

TensorFlow expects separate arguments for features and labels (or a tuple if using multiple inputs/outputs):

model.fit(X_train, y_train, epochs=5)

3. Issues with @tf.function and numpy array parameters #

If you're using TensorFlow's @tf.function decorator, passing numpy arrays can sometimes trigger hash checks inside the function (especially if you're using them in caches or dictionaries).

Example of the error:

import tensorflow as tf
import numpy as np

@tf.function
def predict_with_cache(input_arr):
    prediction_cache = {}
    prediction_cache[input_arr] = model(input_arr)  # Error here
    return prediction_cache[input_arr]

predict_with_cache(np.array([0.1, 0.2]))

Fix it by converting the array to a hashable type inside the function:

@tf.function
def predict_with_cache(input_arr):
    prediction_cache = {}
    # Convert numpy array to a tuple (or use tensor.numpy() if working with tensors)
    cache_key = tuple(input_arr.numpy())
    prediction_cache[cache_key] = model(input_arr)
    return prediction_cache[cache_key]

Quick Debugging Step #

First, look at the full error stack trace—it will tell you exactly which line of code is causing the problem. That's the fastest way to narrow down whether it's a dictionary/set issue, a TensorFlow data passing issue, or something else.

Once you find that line, ask yourself: Am I using this numpy array somewhere that needs a fixed, unchanging value (like a key)? If yes, convert it to a tuple or adjust your logic to avoid using the array in that context.

内容的提问来源于stack exchange，提问作者Taby