Skip to main content

Caching mindful of computation/storage costs

Project description

Caching for Analytic Computations
---------------------------------

Humans repeat stuff. Caching helps.

Normal caching policies like LRU aren't well suited for analytic computations
where both the cost of recomputation and the cost of storge routinely vary by
one milllion or more. Consider the following computations

```python
# Want this
np.std(x) # tiny result, costly to recompute

# Don't want this
np.transpose(x) # huge result, cheap to recompute
```

Cachey tries to hold on to values that have the following characteristics

1. Expensive to recompute (in seconds)
2. Cheap to store (in bytes)
3. Frequently used
4. Recenty used

It accomplishes this by adding the following to each items score on each access

score += compute_time / num_bytes * (1 + eps) ** tick_time

For some small value of epsilon (which determines the memory halflife.) This
has units of inverse bandwidth, has exponential decay of old results and
roughly linear amplification of repeated results.

Example
-------

```python
>>> from cachey import Cache
>>> c = Cache(1e9, 1) # 1 GB, cut off anything with cost 1 or less

>>> c.put('x', 'some value', cost=3)
>>> c.put('y', 'other value', cost=2)

>>> c.get('x')
'some value'
```

This also has a `memoize` method

```python
>>> memo_f = c.memoize(f)
```

Status
------

Cachey is new and not robust.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cachey-0.1.1.tar.gz (6.1 kB view details)

Uploaded Source

File details

Details for the file cachey-0.1.1.tar.gz.

File metadata

  • Download URL: cachey-0.1.1.tar.gz
  • Upload date:
  • Size: 6.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No

File hashes

Hashes for cachey-0.1.1.tar.gz
Algorithm Hash digest
SHA256 de1db64409158d40acdfdd869ccbca7ce90b2f2ba20d44fb68f32e72f5679f21
MD5 c2102283a25e9c1995dc27d31ebb0070
BLAKE2b-256 f8a961f4a40e5a64ff8969c8f1716d0ecdd4309af7a545c2610abbc1034fddae

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page