pathtrees for latest version

I created a new version because the old one had a different pypi install name from the import name so I created one that was consistent across the board - pathtrees. It also has a much cleaner implementation and has less bespoke/obscure functionality

pathtree

Named path hierarchies

The goal of this package is to make it easier to both define and understand folder structures in your scripts and pipelines. I find it difficult to visualize file structure when looking at a series of os.path.join commands scattered throughout a script, so instead, you define all of the paths up top, and fill in the placeholders later.

Install

pip install path-tree

Tutorial

Quick start

NOTE: I want to update this to be more illustrative and less verbose. The problem is that this package can do a lot, but it's hard to show it concisely out of context. I'm not sure when I'll get a chance, but I'll try to get to it!

Also, I apologize for not having an API Reference. I haven't quite ironed that into my deployment pipeline yet!

import pathtree

# instantiate the path tree - dict nesting == folder nesting !
paths = pathtree.tree('./logs', {
    '{log_id}': {
        'model.h5': 'model',
        'model_spec.pkl': 'model_spec',
        'plots': {
            'epoch_{i_epoch:04d}': {
                '{plot_name}.png': 'plot',
                '': 'plot_dir'
            }
        }
    }
})

...

# specify variables along the way as they become available
paths.update(log_id=12345)

# usually they aren't inserted until you call format, so you can still change them
paths.update(log_id=12346)

...

# when you're ready to construct the path, just call .format()
# with the remaining variables and it will give you the formatted
plt.imwrite(paths.plot.format(i_epoch=5, plot_name='f1_score'))
# writes to ./logs/12345/plots/epoch_0005/f1_score.png

The old way - makes my brain sad

Usually, you end up defining paths doing something like this (unless I'm doing something weird/dumb - lmk !!). And often, these end up scattered throughout your project and getting a high level picture of your path hierarchy is difficult.

base_log_dir = './blah/logs'
run_dir = os.path.join(base_log_dir, log_id)
resources_dir = os.path.join(run_dir, 'resources')

...

model_file = os.path.join(run_dir, 'model.h5')
model_spec = os.path.join(run_dir, 'model_spec.pkl')

...

pump_file = os.path.join(resources_dir, 'pump.pkl')

...

plot_dir = os.path.join(run_dir, 'plots', 'epoch_{i_epoch:04d}')
plot_file = os.path.join(plot_dir, '{plot_name}.png')

The new way ! brain smiles

Instead, you can define your path hierarchy all in one place and give each tree node a name.

import os
import pathtree

# define the entire directory structure
# the tree keys represent folder names.
# the final non-dict key represents the name for
# that directory node.
#
# e.g.: {folder1: {folder2: name}}
# paths.name => folder1/folder2
#
# Notice the blank key under "plots". This takes advantage
# of the fact that os.path.join('plots', '') == 'plots'.
# So the name assigned to the blank string is naming the
# directory
base_paths = pathtree.tree('./logs', {
    '{log_id}': {
        'model.h5': 'model',
        'model_spec.pkl': 'model_spec',
        'plots': {
            'epoch_{i_epoch:04d}': {
                '{plot_name}.png': 'plot',
                '': 'plot_dir' # name for the directory
            }
        }
    }
})

# specify the log_id - specify returns a copy, update operates inplace
paths = base_paths.specify(log_id=12345)

Basic Concepts

Paths - a collection of paths items as defined using pathtree.tree. Essentially, it is a wrapper around a flat dictionary of name -> path and a data dictionary that is provided to all paths format.
Path - a single path. It extends os.PathLike so it can be used where path objects are expected (e.g. open(path).read()). It is a wrapper around a pathlib.Path object and a data dictionary. It provides basic path operations (join('subdir'), .up(2) to go up 2 parent directories, .glob() glob replacing missing fields with '*').

Conversion to string

You can access the paths using the name defined in the tree:

assert str(paths.model_spec) == './logs/12345/model_spec.pkl'

A Paths object (as defined above) is really just a dictionary of name => pathtree.Path objects. This is a wrapper around string format pattern and a data dictionary.

To convert a Path to a string, there are a few ways to get what you want.

For fully specified strings (meaning that str.format will run without a KeyError), all three of these methods return the same thing: a fully formatted string.

assert paths.model.format() == './logs/12345/model.h5'
assert paths.model.partial_format() == './logs/12345/model.h5'
assert paths.model.maybe_format() == './logs/12345/model.h5'

For an underspecified string (missing data keys), the return values are different:

# str.format is missing a key and will throw an exception
try:
    paths.plot.format(plot_name='f1_score')
    assert False
except KeyError: # missing i_epoch
    assert True

# str.format is missing a key so the missing key will be left in
assert (paths.plot.partial_format(plot_name='f1_score') ==  
        './logs/12345/plots/epoch_{i_epoch:04d}/f1_score.png')

# str.format is missing a key so it will keep it as a Path object
# with the updated key `plot_name`.
# this retains the ability to use data updating functionality.
assert isinstance(
    paths.plot.maybe_format(plot_name='f1_score'), pathtree.Path)

Paths are castable to string which is the same as partial_format. You can also access the unformatted path using Path().path.

assert str(paths.plot) == paths.plot.partial_format()
assert paths.plot.path == the_unformatted_plot_path

pathtree.Path subclasses os.PathLike, meaning that os.path functions know how to convert it to a path automatically.

assert isinstance(paths.model, os.PathLike)
assert os.path.join(paths.model) == paths.model.format()
assert os.path.isfile(paths.model)

Updating format data

You can add path specificity at various points along the way. This is helpful when you need to reference subdirectories based on some loops or similar pattern. You can do that in a couple ways:

# across the entire directory object

# update in place
paths.update(log_id=12345)

# or create a copy
paths2 = paths.specify(log_id=23456)
assert paths.data['log_id'] == 12345 and paths2.data['log_id'] = 23456

# reverse specify - remove a data key
paths2 = paths2.unspecify('log_id')
assert 'log_id' not in paths2.data

# or for a single path

# in place
plot_file = paths.plot
plot_file.update(plot_name='f1_score')

# create a copy
plot_file = paths.plot.specify(plot_name='f1_score')

assert 'plot_name' not in paths.data
assert plot_file.data['plot_name'] == 'f1_score'

# reverse specify - remove a data key
plot_file = plot_file.unspecify('plot_name')
assert 'plot_name' not in plot_file.data

Additional Features

You can automatically do glob searching. Any missing fields will be filled with a glob wildcard (asterisk). Note that this would fail using plain string format because the leading zero formatter (:04d) will throw an error if you try to insert '*' (because it's a string).

plot_file = paths.plot.specify(plot_name='f1_score')

assert (plot_file.partial_format() ==
        './logs/12345/plots/epoch_{i_epoch:04d}/f1_score.png')
assert (plot_file.glob_pattern ==
        './logs/12345/plots/epoch_*/f1_score.png')
assert (plot_file.glob() ==
        glob.glob('./logs/12345/plots/epoch_*/f1_score.png'))

You can also sometimes parse out data from a formatted string. Be warned, it may not always work correctly because sometimes the parsing is ambiguous. See https://github.com/r1chardj0n3s/parse#potential-gotchas

plot_file = paths.plot.specify(root='some/logs')

assert (plot_file.partial_format() ==
        'some/logs/12345/plots/epoch_{i_epoch:04d}/{plot_name}.png')

expected = {
    'root': 'some/logs'
    'log_id': '12345',
    'i_epoch': '0002',
    'plot_name': 'f1_score.png',
}


plot_data = plot_file.parse('./logs/12345/plots/0002/f1_score.png')
assert set(plot_data.items()) == set(expected.items())

beasteers/pathtree