Claude gave $250 for Claude code using through web for two weeks, you visit claude.ai/code to connect your github account, but you have to installClaude GitHub app in your repo first, otherwise the connect will not work. Then you will need to set up which repo you want claude to have read/write access. Claude will pull the repo to a Anthropic-managed virtual machine, and execute code there.

The testing was done based on https://github.com/someonebio/deephyper, I asked Claude code to add an example, and it did: https://github.com/deephyper/deephyper/compare/master…someonebio:deephyper:claude/deephyper-benchmark-example-011CUoRaMUPsoW2QHcz2QpfF. It added the commit, and wrote all the code. Cost me $1.

Then I verified it on local machine use Claude to run based on TESTING_GUIDE.md, it solved the environment issue, and give some solution to install pymoo since python/13.4 is too new for pymoo/0.6.1.5. So it Downloaded the pre-compiled conda package for pymoo 0.6.1.5 and install it manually. The code is finally runnable, the quick test pass, but the full benchmark did not finish with segmentation error, which seems to be image gereration related. I post the error to Claude, and it fixed it by editing the code locally.

Test Results Summary:

Test 1: Imports ✓

All required DeepHyper modules imported successfully:

  • numpy
  • deephyper.hpo (HpProblem, CBO, RandomSearch)
  • deephyper.evaluator (Evaluator)
  • deephyper.evaluator.callback (TqdmCallback)
  • deephyper.analysis.hpo (parameters_at_max)

Test 2: Ackley Function ✓

  • At origin [0,0,0,0,0]: 0.0000000000 (correct, expected ~0)
  • At [1,1,1,1,1]: 3.625385

Test 3: Problem Definition ✓

  • Successfully created HpProblem with 5 hyperparameters
  • Each hyperparameter range: (-32.768, 32.768)

Test 4: Run Function ✓

  • Run function defined successfully
  • Properly extracts parameters and evaluates Ackley function
  • Returns negated value for maximization

Test 5: Quick Optimization Test ✓

  • Evaluations: 5 (completed successfully)
  • Best objective found: -20.122708
  • Best configuration: [-3.3041, -11.1007, 18.9604, 28.0214, -11.1129]
  • Results DataFrame contains required columns: objective, job_id

Installation Challenges and Solutions

Challenge 1: Missing Dependencies

Issue: Multiple Python packages not installed (numpy, matplotlib, scikit-learn, pandas, etc.)

Solution:

python3 -m pip install numpy matplotlib scikit-learn pandas tqdm ConfigSpace
python3 -m pip install pydantic loky pyyaml psutil parse dill cloudpickle

Challenge 2: pymoo Build Failure (Critical)

Issue: pymoo>=0.6.0 required but failed to build from source due to C++ compiler errors:

  • Error: 'ios' file not found - missing C++ standard library headers
  • Python 3.14 is very new; pymoo compilation not yet fully supported
  • Tried multiple approaches:
    • Installing with CFLAGS/CXXFLAGS
    • Using different SDK paths
    • Installing Cython

Solution: Used pre-compiled conda package

# Install zstd for extraction
brew install zstd

# Download pre-compiled pymoo from conda-forge
curl -L "https://conda.anaconda.org/conda-forge/osx-arm64/pymoo-0.6.1.5-py314ha3d490a_1.conda" \
  -o /tmp/pymoo.conda

# Extract conda package
cd /tmp
tar -xf pymoo.conda
/opt/homebrew/bin/zstd -d pkg-pymoo-0.6.1.5-py314ha3d490a_1.tar.zst
tar -xf pkg-pymoo-0.6.1.5-py314ha3d490a_1.tar

# Copy to Python site-packages
cp -r /tmp/lib/python3.14/site-packages/pymoo \
  /Users/someone/.pixi/envs/python/lib/python3.14/site-packages/

# Install pymoo runtime dependencies
python3 -m pip install alive-progress Deprecated

Verification:

import pymoo
print(pymoo.__version__)  # Output: 0.6.1.5

Environment Details

  • Architecture: ARM64 (Apple Silicon)
  • Python Version: 3.14.0
  • Python Location: /Users/someone/.pixi/bin/python3
  • Git Branch: claude/deephyper-benchmark-example-011CUoRaMUPsoW2QHcz2QpfF

Key Dependencies Installed

Package Version Source
numpy 2.3.4 PyPI
matplotlib 3.10.7 PyPI
scikit-learn 1.7.2 PyPI
pandas 2.3.3 PyPI
scipy 1.16.3 PyPI
ConfigSpace 1.2.1 PyPI
pydantic 2.12.3 PyPI
loky 3.5.6 PyPI
pymoo 0.6.1.5 conda-forge
alive-progress 3.3.0 PyPI
Deprecated 1.3.1 PyPI

Running the Tests

Quick Test (5 evaluations)

PYTHONPATH=/Users/someone/projects/deephyper/src:$PYTHONPATH \
  python3 test_benchmark.py

Expected time: ~5-10 seconds Result: ✅ PASSED

Full Benchmark (100 evaluations with visualizations)

PYTHONPATH=/Users/someone/projects/deephyper/src:$PYTHONPATH \
  MPLBACKEND=Agg \
  python3 examples/examples_bbo/plot_benchmark_ackley_optimization.py

Expected time: ~1-3 minutes Results: ❌ Failed with error.. —

API compatibility issue

  1. Fixed the bug in examples/examples_bbo/plot_benchmark_ackley_optimization.py:277 by removing the invalid show=False parameter from the plot_search_trajectory_single_objective_hpo() function call. 
    PYTHONPATH=/Users/someone/projects/deephyper/src:$PYTHONPATH MPLBACKEND=Agg python3 examples/examples_bbo/plot_benchmark_ackley_optimization.py

    Result: ✅ PASSED —

A notable issue is this make generating new PRs on github so easy that you do not need to know git or python or Ackley(I have no idea what it is). This function only supports for github so far, not gitlab (yet).

<
Previous Post
Single-Cell QC Analysis Summary from claude skill
>
Next Post
Single-Cell RNA-seq Analysis using K-dense open source Claude skill.