Qiskit 2.0环境下BaseEstimator导入错误及版本兼容问题求助

问题背景 #

当前conda环境（/opt/anaconda3/envs/qnn）内Qiskit相关包版本：

conda list qiskit
# packages in environment at /opt/anaconda3/envs/qnn:
#
# Name                    Version                   Build  Channel
qiskit                    2.0.0           py312hcd83bfe_0    conda-forge
qiskit-aer                0.17.0                   pypi_0    pypi
qiskit-ibm-runtime        0.37.0                   pypi_0    pypi
qiskit-machine-learning   0.8.2              pyhd8ed1ab_1    conda-forge

执行导入代码时：

import numpy as np
import pandas as pd
from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
from qiskit.quantum_info import SparsePauliOp
from qiskit.primitives import StatevectorEstimator as Estimator
from qiskit_machine_learning.neural_networks import EstimatorQNN

出现以下错误：

ImportError:
----> 8 from qiskit_machine_learning.neural_networks import EstimatorQNN

[...]

File /opt/anaconda3/envs/qnn/lib/python3.12/site-packages/qiskit_machine_learning/neural_networks/effective_dimension.py:23
...
---> 24 from qiskit.primitives import BaseEstimator, BaseEstimatorV1, Estimator, EstimatorResult
     25 from qiskit.quantum_info import SparsePauliOp
     26 from qiskit.quantum_info.operators.base_operator import BaseOperator

ImportError: cannot import name 'BaseEstimator' from 'qiskit.primitives' (/opt/anaconda3/envs/qnn/lib/python3.12/site-packages/qiskit/primitives/__init__.py)

单独执行from qiskit.primitives import Estimator也报错：

ImportError: cannot import name 'Estimator' from 'qiskit.primitives' (/opt/anaconda3/envs/qnn/lib/python3.12/site-packages/qiskit/primitives/__init__.py)

尝试降级Qiskit到0.44.0未解决，通过conda升级qiskit-machine-learning到0.9.0时出现包未找到错误：

conda install -c conda-forge qiskit-machine-learning=0.9.0
Channels:
 - conda-forge
 - defaults
Platform: osx-arm64
Collecting package metadata (repodata.json): done
Solving environment: failed

PackagesNotFoundError: The following packages are not available from current channels:
  - qiskit-machine-learning=0.9.0*

解决方案 #

1. 核心原因说明 #

Qiskit 2.0.0是重大版本更新，重构了primitives模块：原Estimator被拆分为StatevectorEstimator、SamplerEstimator等具体实现，且移除了BaseEstimator这类V1版本的抽象类。而qiskit-machine-learning 0.8.2依赖的是Qiskit 0.4x系列的旧primitives结构，因此出现兼容问题。

qiskit-machine-learning 0.9.0才适配Qiskit 2.0.0，但conda-forge的osx-arm64通道暂未提供该版本，需改用pip安装。

2. 具体操作步骤 #

• 步骤1：卸载现有冲突包
  先移除当前的Qiskit核心包和机器学习包，避免版本残留：

  conda remove qiskit qiskit-machine-learning -y
  pip uninstall qiskit-aer qiskit-ibm-runtime -y
  

• 步骤2：安装适配的版本组合
  有两种稳定组合可选：

  • 组合一：Qiskit 2.0.0 + qiskit-machine-learning 0.9.0（推荐）
    用conda安装核心Qiskit，pip安装适配的机器学习扩展包：

    conda install -c conda-forge qiskit=2.0.0 -y
    pip install qiskit-machine-learning==0.9.0 qiskit-aer==0.18.0 qiskit-ibm-runtime==0.38.0
    

    注：qiskit-aer 0.18.0和qiskit-ibm-runtime 0.38.0是适配Qiskit 2.0.0的对应版本。

  • 组合二：Qiskit 0.44.1 + qiskit-machine-learning 0.8.2（兼容旧代码）
    如果不想升级到Qiskit 2.0，需确保所有依赖包版本同步匹配：

    conda install -c conda-forge qiskit=0.44.1 qiskit-machine-learning=0.8.2 -y
    pip install qiskit-aer==0.17.1 qiskit-ibm-runtime==0.37.1
    

    注：之前降级到0.44.0未解决问题，大概率是因为qiskit-aer、qiskit-ibm-runtime等依赖包版本未同步降级导致的。

• 步骤3：验证导入
  重新运行导入代码，确认无错误：

  import numpy as np
  import pandas as pd
  from qiskit import QuantumCircuit
  from qiskit.circuit import Parameter
  from qiskit.quantum_info import SparsePauliOp
  # 对于Qiskit 2.0，Estimator需从具体实现导入，或用别名
  from qiskit.primitives import StatevectorEstimator as Estimator
  from qiskit_machine_learning.neural_networks import EstimatorQNN
  

3. 后续注意事项 #

• 尽量避免混用conda和pip安装Qiskit相关包，优先保持同一安装源；若必须混用，建议用conda安装核心Qiskit包，pip安装扩展包。
• 安装前确认qiskit-machine-learning的版本兼容要求，确保与Qiskit核心版本匹配。

内容的提问来源于stack exchange，提问作者diogomaia00