Introduction

InvalidArgumentError: Expected dimension in the range [-1, 1) but got 1 usually means you passed an axis that does not exist for the tensor you are working with. In plain terms, your code is treating a one-dimensional tensor like it had a second dimension.

What the Error Range Means

Tensor operations often take an axis argument. TensorFlow validates that axis against the tensor rank.

If the allowed range is [-1, 1), the only valid axes are:

• '-1, meaning the last axis'
• '0, meaning the first axis'

Axis 1 is invalid because a rank-1 tensor has only one dimension. There is no second axis.

For example, this tensor:

1import tensorflow as tf
2
3x = tf.constant([10, 20, 30])
4print(x.shape)

has shape (3,), so it is rank 1. Valid axes are only 0 and -1.

A Typical Way to Trigger the Error

One common mistake is trying to concatenate one-dimensional tensors along axis 1:

1import tensorflow as tf
2
3a = tf.constant([1, 2, 3])
4b = tf.constant([4, 5, 6])
5
6tf.concat([a, b], axis=1)

This fails because a and b are vectors, not matrices. They do not have an axis 1.

The correct version for vectors is:

result = tf.concat([a, b], axis=0)
print(result)

If you truly want to concatenate across columns, first reshape them into rank-2 tensors:

1a2 = tf.reshape(a, (3, 1))
2b2 = tf.reshape(b, (3, 1))
3
4result = tf.concat([a2, b2], axis=1)
5print(result)

How to Debug It Quickly

Whenever you see this error, inspect the tensor shape before the failing operation:

print(x.shape)
print(tf.rank(x))

Then compare that rank to the axis you passed. If rank is 1, axis 1 is invalid. If rank is 2, axis 1 is fine. This simple check usually reveals the problem faster than reading the full stack trace.

It is also useful to add assertions in data pipelines:

tf.debugging.assert_rank(x, 2)

That gives you a clearer failure earlier in the pipeline if your code expects a matrix but receives a vector.

Why This Happens in Model Code

This error often appears when preprocessing steps accidentally drop a dimension. A few common examples:

• selecting one item from a batch and losing the batch axis
• using tf.squeeze() too aggressively
• reading labels as shape (batch,) while the model expects (batch, 1)
• flattening tensors earlier than intended

Consider this example:

1features = tf.constant([[1.0, 2.0], [3.0, 4.0]])
2single_row = features[0]
3
4print(features.shape)   # (2, 2)
5print(single_row.shape) # (2,)

After indexing with features[0], the result is rank 1. If later code still assumes a two-dimensional tensor and uses axis 1, the error appears.

Fixing It the Right Way

The fix depends on what shape you actually want.

If the data should stay one-dimensional, use axis 0:

values = tf.constant([1, 2, 3])
print(tf.reduce_sum(values, axis=0))

If the data should be two-dimensional, add or preserve the missing dimension:

values = tf.constant([1, 2, 3])
column = tf.expand_dims(values, axis=1)
print(column.shape)

Now axis 1 exists because the shape becomes (3, 1).

For labels, many training bugs are fixed by being explicit:

y = tf.constant([0, 1, 1, 0], dtype=tf.float32)
y = tf.reshape(y, (-1, 1))
print(y.shape)

This turns a rank-1 label vector into a column-shaped rank-2 tensor.

Build Shape Awareness into Your Workflow

Tensor bugs are easier to prevent than to debug. A few habits help a lot:

• print shapes at the boundaries of your pipeline
• keep batch dimensions explicit
• use reshape, expand_dims, and squeeze deliberately
• read API docs for each operation's allowed axis range

Once you start thinking in ranks and shapes, this error message becomes much less mysterious.

Common Pitfalls

• Assuming a vector has axis 1 when it only has axis 0.
• Losing a dimension through indexing or squeeze() and not noticing.
• Trying to fix the error by changing the axis randomly instead of checking the intended shape.
• Reshaping blindly without confirming that the new shape matches the model's expectations.
• Debugging only the failing line instead of the earlier preprocessing step that changed the rank.

Summary

• This TensorFlow error means the axis argument is out of bounds for the tensor rank.
• A range of [-1, 1) means only axes -1 and 0 are valid.
• The most common cause is treating a rank-1 tensor like a rank-2 tensor.
• Check tensor shapes before the failing operation and preserve dimensions intentionally.
• Use reshape or expand_dims when you truly need an extra axis.