Squash Pickle

Like a pickle, only smaller*.

Tiny python package that compresses your pickles using gzip. Quacks like a pickle.

* For small objects (< 100 bytes) gzip overhead can end up increasing in size. Only squash your pickles when you are working with big objects.

---

Getting Started

First install the package, this has no additional dependencies:

pip install squashpickle

Then replace your pickle calls with squashpickle ones. squashpickle implements, dump, dumps, load, and loads functions.

To avoid confusion, it is recommended to dump and load to file with .gz extension eg: pkl.gz, but there is nothing enforcing this.

---

Performance

The GZIP compression can have a HUGE impact on large objects. Say you are pickling something like a polars / pandas dataframe; these pickles may end up being hundreds of MBs. With squashpickle can get compression ratios exceeding 10x.

For example, if we load a large dataframe of Australian weather data. Using pickle, this object serializes to 37794198 bytes (~37.8MB). Dumping the same dataframe with squashpickle results in 3370363 bytes (~3.4MB), around 9% of the overall file.

import polars as pl
import pickle
import squashpickle

df = pl.read_csv(r"C:\temp\weatherAUS.csv", null_values=["NA"])
print(len(pickle.dumps(df)), len(squashpickle.dumps(df)))

As with any compression, there is a performance cost to achieving the smaller files. For objects <1MB this is hardly noticeable, but for objects hundreds of MBs the delay can be significant. It'll depend on your use case if this is a worthwhile tradeoff.

---

Security

squashpickle uses Python's pickle and gzip modules internally. As with pickle and gzip, never load data from untrusted sources. Deserializing malicious pickle data can execute arbitrary code.