ModuleNotFoundError fastparquet

Not specified by the original author.

TL;DR: pandas needs a parquet engine and fastparquet is the one your code asked for. Run pip install fastparquet (pyarrow works too) and the read succeeds.

The error

ModuleNotFoundError: No module named 'fastparquet'

Fix it

1. Install an engine:

   pip install fastparquet

   Expected: install completes.

2. Verify:

   import pandas as pd
   df = pd.read_parquet("data.parquet", engine="fastparquet")
   print(df.head())

   Expected: prints the first rows.

3. Alternative engine: pip install pyarrow and omit the engine argument.

When this applies

Use this when readparquet or toparquet raises the fastparquet error.

When it does not apply

If the error is the missing-pyarrow message, install pyarrow instead. If an engine is installed but the file will not read, the file may be corrupt.

Compatibility

All pandas versions with parquet support.

Root cause

pandas delegates parquet to an external engine and ships with none selected by default. Code (or pandas' own engine preference) asking for fastparquet fails when it was never installed.

Edge cases

pyarrow is the recommended engine for new code; fastparquet stays useful for older files. In notebooks, restart the kernel after installing so the new package is visible.

GitHub issue pandas-dev/pandas#18156 reports ModuleNotFoundError: No module named 'fastparquet' when using pandas parquet IO without an engine installed. The verified fix: installing an engine, either pip install fastparquet or pip install pyarrow, after which read_parquet works. Full thread: https://github.com/pandas-dev/pandas/issues/18156

Not specified by the original author.

TL;DR: fastparquet's compiled speed-up module did not build during install. Reinstall fastparquet with a working C compiler (or grab a prebuilt wheel) and the import works.

The error

ModuleNotFoundError: No module named 'fastparquet.cencoding'

Fix it

1. Reinstall fastparquet, letting pip pick a prebuilt wheel:

   pip install --force-reinstall fastparquet

   Expected: installs a wheel that already contains cencoding.

2. If pip builds from source instead, install a C compiler first (Debian/Ubuntu: sudo apt install build-essential; then rerun the reinstall). Expected: the build compiles cencoding.

3. Verify:

   python -c "import fastparquet; print(fastparquet.__version__)"

   Expected: prints the version, no error.

When this applies

Use this when import fastparquet raises the cencoding error, typical on platforms without a matching wheel.

When it does not apply

If the error is plain No module named 'fastparquet', just install it. If cencoding imports but reads are wrong, that is a data issue, not this one.

Compatibility

fastparquet installs that compile the optional C extension; wheels for major platforms already include it.

Root cause

fastparquet ships an optional Cython extension (cencoding) for speed. Source installs without a compiler silently skip it, and the import of the extension module then fails.

Edge cases

fastparquet works in pure-python mode without cencoding, just slower; the import error only bites because the package tries the extension first. Pinning fastparquet in CI avoids surprise source builds.

GitHub issue dask/fastparquet#977 reports ModuleNotFoundError: No module named 'fastparquet.cencoding' when importing fastparquet, because the C speed-up extension failed to compile during install. The verified fix: reinstalling fastparquet with a working C compiler toolchain (or installing a prebuilt wheel that already contains cencoding) so the extension builds, after which the import succeeds. Full thread: https://github.com/dask/fastparquet/issues/977

Not specified by the original author.

TL;DR: your fastparquet is too old for your pandas. pandas removed RangeIndex from pandas.core.index, so upgrade fastparquet to a compatible release, or pin an older pandas.

The error

ImportError: cannot import name RangeIndex

Fix it

1. Upgrade fastparquet:

   pip install -U fastparquet

   Expected: installs a release that imports RangeIndex from the right place.

2. Verify:

   python -c "import fastparquet; print('ok')"

   Expected: imports cleanly.

3. Workaround if you cannot upgrade fastparquet: pin pandas to a version that still has the old path:

   pip install "pandas<1.0"

   Expected: the old import path works again.

When this applies

Use this when import fastparquet raises the RangeIndex error with a modern pandas.

When it does not apply

If your own code imports RangeIndex from pandas.core.index, use pandas.RangeIndex instead. If fastparquet is current, reinstall it.

Compatibility

Old fastparquet with pandas 1.0+. pandas.RangeIndex is the stable public path on all pandas versions.

Root cause

pandas reorganized its internals and removed the pandas.core.index alias for RangeIndex. fastparquet imported the internal path, so the removal broke it at import time.

Edge cases

Pinning ancient pandas breaks most of the modern stack; upgrading fastparquet is strongly preferred. The same pandas-internal cleanup broke other old libraries the same way.

GitHub issue dask/fastparquet#269 reports ImportError: cannot import name RangeIndex when importing fastparquet, with the traceback showing from pandas.core.index import RangeIndex, Index. The verified answer: newer pandas removed RangeIndex from pandas.core.index, so old fastparquet broke; the fix is upgrading fastparquet to a release compatible with the installed pandas (or pinning an older pandas as a workaround). Full thread: https://github.com/dask/fastparquet/issues/269