Basins

Motivation

Basins are a powerful concept that allow you to save time, disk space, and network usage when processing DC data. The basic idea behind basins is that you avoid duplicating feature data by not copying all of the features from an input file to an output file, but by linking from the output file to the input file. Feature data “flows” from the basins to the files downstream.

Due to the fact that basins are implemented in dclab, all software that relies on dclab for opening data files (e.g. DCscope or CytoPlot) automatically supports basins as well.

Use cases

To illustrate how you can use basins in your analysis pipeline, let’s consider the three examples shown in Fig. 1.

  1. You have found a dataset on figshare.com that you would like to work with. However, you are only interested in a fraction of the events in that file. You decide to open the URL of that dataset with dclab, apply a few filters and export the resulting subset to a 500 MB file on disk with basins=True (see Storing the basin information below for exporting with basins). The original file on figshare.com is now a basin of the file on your hard disk. If you e.g. decided not to export image data to your local file, you will still be able to access the images through the basin that is defined in the exported file as long as you have a working internet connection.

  2. You have an automated data analysis pipeline that relies entirely on DCOR. You are uploading your raw data to DCOR. A background job performs segmentation and feature extraction and the resulting data are stored on DCOR as well. To have fast scalar feature access, you download the condensed dataset from DCOR to your computer. The raw image data as well as the background image data and the event masks are accessed via their respective basins when you open the condensed file with dclab. In this use case, you are trading local disk space for slower access to the image and mask features limited by the bandwidth of your internet connection.

  3. Let’s say your pipeline is designed to compute a new feature userdef1 and you would like to open the output file in DCscope, visualizing this feature in combination with other features defined in the input file (e.g. deform). What you could do is write the userdef1 feature directly to the input file or create a copy of your input file and write userdef1 to that copy. However, this might make you feel uneasy, because you would like to avoid editing your original raw data (possible data loss) and copying an entire file seems like unnecessary overhead (redundant data, high disk usage). Furthermore, your raw data are located on a network share in your institute and you want to avoid causing a lot of traffic. The solution with basins is to open the raw input file, run your analysis and store the scalar features created by your analysis on your local computer. With basins, these local files know that the image data are stored on the network share. You have saved yourself the trouble of copying large files across the network.

../_images/basin_example_workflows.svg

Fig. 1 Three exemplary workflows where basins are used. (A) There is one basin consisting of an .rtdc file uploaded to figshare. The user exported partial feature data to their computer. Other feature data are still accessible via the online basin. (B) The user has a DCOR-based analysis pipeline where the raw and processed data are stored on DCOR. The user downloads a condensed version of the data and can access image data via basins. (C) The raw image data are located on a network share at the user’s institute. The user writes one script to e.g. extract the relevant events (processed 1) and a second script to e.g. compute additional features (processed 2).

To summarize the advantages of using basins:

  • Avoid data redundancy. Since basins defined in basins are also just basins, you can design your analysis pipeline as a chain of basins. The raw file contains the image data and subsequent steps in the pipeline only add features or gate events. You never have to store the same feature twice on disk.

  • Work with read-only datasets. When you are computing new features, you do not have to modify any existing files on disk and can instead link to them. This means you can have your data on read-only file storage (e.g. online data) and don’t have to fear accidental data loss.

  • Fast data access. You do not have to download an entire dataset in order to obtain a simple scatter plot for offline usage. Instead, you can download only a few relevant scalar features, perform your analysis, and afterwards you even have the option to look at the images in the basin.

Definitions

To clarify the notation, please read these definition of the terms used in dclab in the context of basins. They also serve as a complete introduction to the capabilities of basins.

referrer:

A dataset that defines a basin “refers” to that basin. If a dataset is a referrer, then the dataset has features stored in other datasets (the basins).

basin:

File or remote location containing feature data that are made accessible to the referrer. Both the referrer and the basin must originate from the same DC measurement. E.g. You can work with a local .rtdc file containing only :ref:´scalar features <sec_features_scalar>` with the image data stored in a basin on DCOR to keep your local disk usage low.

mapped basin:

A mapped basin is a basin that does not share the same event indices with its referrer. This means that either

  • the basin is a superset of the referrer: This happens most often when you export a subset of events to a file, resulting in a referrer that has less events than the (mapped) basin file. This case is useful when you only need a subset of features in the referrer, but don’t want to (spend the time to) store all feature data (e.g. image data) in the referrer to save disk space.

  • the basin is a subset of the referrer: This can happen when you analyze your data with ChipStream. For every image in the basin, the referrer defines one or more events, resulting in a referrer that has more events than the basin.

  • both cases above at the same time: There is really no limit in defining basins. You can have two referrer events map to one basin event or the other way around. The only limiting factor is that both, basin and referrer, must be derived from the same original measurement.

Mapped basins allow you to minimize data redundancy for analysis pipelines that produce output files (referrers) with a subset of the events from the input file (basin).

Note

To be able to map from the input file to the output file, dclab stores the mapping information as integer indices in dedicated features enumerated basinmap0, basinmap1, etc.

internal basin:

This is a special basin type, developed to reduce disk usage for background images (image_bg). An internal basin stores the basin data within the referrer file. While unintuitive for regular feature data, using internal basins to store background image data, with one background image per second, can significantly reduce disk usage due to the many-to-one-mapping nature of the problem.

In addition, let’s digest the following definitions, which are also keyword arguments to the RTDCWriter.store_basin method used further below.

basin type:

A basin can be a local file (including files on a network share), an internal basin (see above), or a remote file which means that it is accessible via a networking protocol. Local basins can be defined either via absolute and/or relative paths. Remote basins can be simple links (e.g. download links for a figshare resource), DCOR resource identifiers, or links to an object in an S3-compatible object store (e.g. Amazon S3 or OpenStack Swift).

basin format:

This is the subclass of RTDCBase that defines how the basin is accessed. For file-type basins, this is “hdf5” and for remote-type basins, this is “dcor”, “http”, or “s3”.

basin mapping:

If the events enumerated in the referrer are identical to the events in the basin, then we call the mapping “same”. Otherwise, we call it a mapped basin followin the definition above.

basin features:

Feature information in a dataset as defined in the general feature section. Basins are only ever defined for features. There is no such thing as basins for metadata, tables, or logs. You may define basins and explicitly state the features this basin provides. In combination with mapping, you could e.g. realize your own event segmentation pipeline, storing only the mask feature and extracted scalar features in you output file, while you define the image feature via the input file basin. If you combine this approach with the dcor basin format, you can distribute all of your data (raw and processed) in a very efficient and transparent manner.

Note

Note that basins are locations upstream in your analysis pipeline. Features flow from basins downstream to the referrers. When a dataset has basins, this means that there are other files (the basins) that contain additional feature data.

Note

Basins can have basins. A referrer can refer to multiple basins. And a referrer can be a basin as well. Basins of basins are passed down to referrers downstream.

These definitions should already give you a good feeling about how you can employ basins in your workflow. As a final note, be aware that you can also define basins recursively. Basins can have basins. And dclab has a check for circular basin definitions so you don’t have to worry about that as well.

Defining Basins

Basins may have different properties depending on the use case. Let’s dive into an example:

import dclab

with (dclab.new_dataset("input.rtdc") as ds,
      dclab.RTDCWriter("output.rtdc") as hw):
    # `ds` is the basin
    # `hw` is the referrer

    # First of all, we have to copy the metadata from the input file
    # to the output file. If we forget to do this, then dclab will
    # not be able to open the output file.
    hw.store_metadata(ds.config.as_dict(pop_filtering=True))

    # Next, we can compute and write the new feature to the output file.
    hw.store_feature("userdef1", np.random.random(len(ds)))

    # Finally, we write the basin information to the output file.
    hw.store_basin(
        basin_name="raw data",
        basin_type="file",
        basin_format="hdf5",
        basin_locs=["input.rtdc"],
    )

# You can now open the output file and verify that everything worked.
with dclab.new_dataset("output.rtdc") as ds_out:
    assert "userdef1" in ds_out, "check that the feature we wrote is there"
    assert "image" in ds_out, "check that we can access basin features"
    # You could also be more specific:
    assert "userdef1" in ds_out.features_innate
    assert "image" in ds_out.features_basin

What happened? First, we created an output.rtdc file that contains the metadata from the ìnput.rtdc file. This is important so that dclab can verify the basin when we open the referrer. Then, we wrote the feature userdef1, filled with random data, to the referrer. Finally we stored the basin information referencing all features from the input.rtdc file. To make sure everything worked, we opened the output referrer file and saw that dclanb transparently gives us access to the features stored in the referrer and the basin.

Examples

Mapped basin via RTDCWriter

You can explicitly define a mapped basin via the RTDCWriter.store_basin method (see also the example after this one).

import dclab
import numpy as np

with (dclab.new_dataset("input.rtdc") as ds,
      dclab.RTDCWriter("output.rtdc") as hw):

    # metadata
    hw.store_metadata(ds.config.as_dict(pop_filtering=True))

    # take every second event from the input file
    event_mapping = np.arange(len(ds), None, 2, dtype=np.uint64)

    # write the basin
    hw.store_basin(
        basin_name="raw data",
        basin_type="file",
        basin_format="hdf5",
        basin_locs=["input.rtdc"],
        basin_map=event_mapping,
    )

# verify that this worked
with (dclab.new_dataset("input.rtdc") as ds_in,
      dclab.new_dataset("output.rtdc") as ds_out):
    assert np.allclose(ds_in["deform"][::2], ds_out["deform"])

Implicitly mapped basin via HDF5 export

It is also possible to implicitly write basin information to an exported file, achieving the same result as above (a very small output file).

import dclab
import numpy as np

with dclab.new_dataset("input.rtdc") as ds:
    # remove every second event
    ds.filter.manual[1::2] = False
    ds.apply_filter()
    # export the dataset with the mapped basin
    ds.export.hdf5(path="output.rtdc",
                   features=[],
                   filtered=True,
                   basins=True)

# verify that this worked
with (dclab.new_dataset("input.rtdc") as ds_in,
      dclab.new_dataset("output.rtdc") as ds_out):
    assert np.allclose(ds_in["deform"][::2], ds_out["deform"])

Rewriting Basins

In some situations, you might have to modify the location of a basin, e.g. because you need to make the basins available on different operating systems or because the network share location changed. In those cases, the best approach is to read the basin information, update the basin location and write the updated basin information to that file.

First, locate the basin you would like to modify by listing all basin locations.

with dclab.new_dataset("data_file.rtdc") as ds:
    for ii, bn_dict in enumerate(ds.basins_get_dicts()):
        print(ii, bn_dict["type"], bn_dict.get("paths"), bn_dict["features"])

This will return something like this:

0 file ['/ptmp/data/RC/Reference/2025-02-09_09.46_M003_Reference_5000.rtdc'] ['image']
1 internal ['basin_events'] ["image_bg"]

Note

The second basin in this example (“basin_events” location) is an internal basin (see definitions above).

As you can see, the basin containing the image data is located on a posix path /ptmp which is not accessible on Windows. Assuming you had the same network location mounted on drive P:\\, you could add an additional basin to the file like so:

import json

from dclab.util import hashobj

with dclab.new_dataset("data_file.rtdc") as ds:
    # we want to edit the first file-based basin dictionary containing the image data
    bn_dict = ds.basins_get_dicts()[0]

# replace the path to the basin with the new path
bn_dict["paths"] = [r"P:\\data\RC\Reference\2025-02-09_09.46_M003_Reference_5000.rtdc"]
# remove the "key" from the dictionary (it is part of the old basin)
bn_dict.pop("key")
# convert the basin information to a JSON string
b_lines = json.dumps(bn_dict, indent=2, sort_keys=True).split("\n")
# compute the new basin key
key = hashobj(b_lines)

# write the new basin
with dclab.RTDCWriter("data_file.rtdc") as hw:
    if key not in hw.h5file["basins"]:
        hw.write_text(hw.h5file["basins"], key, b_lines)

After that, you can open the dataset on Windows and access the information in the basin via the mounted network share on drive P:\\.

Accessing private basin data

DCOR

If you have basins defined that point to private data on DCOR, you have to register your DCOR access token in dclab via the static method dclab.rtdc_dataset.fmt_dcor.api.APIHandler.add_api_key().

S3

For basins that point to files on S3, you have to specify the environment variables DCLAB_S3_ACCESS_KEY_ID and DCLAB_S3_SECRET_ACCESS_KEY, and optionally the DCLAB_S3_ENDPOINT_URL as described in the S3 access section.

Basin internals

Storing the basin information

In the output.rtdc file, the basin is stored as a json-encoded string in an HDF5 dataset in the "/basins" group. For the HDF5 export example above, the json data looks like this:

{
  "description": "Exported with dclab 0.58.0",
  "format": "hdf5",
  "name": "Exported data",
  "type": "file",
  "features": null,
  "mapping": "basinmap0",
  "identifier": "1231ae-31f23-342-232-42b1c",
  "paths": [
    "/absolute/path/to/input.rtdc",
    "input.rtdc"
  ]
}

The description and name are filled automatically by dclab here. As expected, the type of the basin is file and the format of the basin is hdf5. There are a few things to notice:

  • The features are set to null which means None, i.e. all features from the input file are allowed.

  • The mapping key reads basinmap0. This is the name of the feature in which to find the mapping information from the input file to the output file. The information can be found in the HDF5 dataset /events/basinmap0 in the output file. Note that the fact that this mapping information is stored as a feature means that it is also properly gated when you define basins iteratively.

  • The identifier is a string that matches the identifier of the dataset. When creating basins without a “same” mapping (as in this case), then the referrer will obtain an identifier that starts with this identifier, but contains additional text. This means identifiers are effectively cryptic data analysis trackers.

  • There are two paths defined, an absolute path (from the root of the file system) and a relative path (relative to the directory of the output file). This relative path makes it possible to copy-paste these two files together to other locations. You will always be able to open the output file and see the basin features defined in the input file. Internally, dclab also checks the run identifier of the output file against that of the input file to avoid loading basin features from the wrong file.

For the sake of completeness, let’s see how the basin information looks like when you derive the output file from a DCOR resource:

import dclab
import numpy as np

with dclab.new_dataset("fb719fb2-bd9f-817a-7d70-f4002af916f0") as ds:
    ds.filter.manual[1::2] = False
    ds.apply_filter()
    ds.export.hdf5(path="output.rtdc",
                   features=[],
                   filtered=True,
                   basins=True)

The corresponding json data:

{
  "description": "Exported with dclab 0.58.0",
  "format": "dcor",
  "name": "Exported data",
  "type": "remote",
  "features": null,
  "mapping": "basinmap0",
  "urls": [
    "https://dcor.mpl.mpg.de/api/3/action/dcserv?id=fb719fb2-bd9f-817a-7d70-f4002af916f0"
  ]
}

As you can see, paths is replaced by urls and the format and type keys changed. The rest remains the same. This also works with private DCOR resources, given that you have globally set your API token as described in the DCOR section.

Basin loading procedure

When dclab opens a dataset the defines a basin, the basin features are retrieved only when they are needed (i.e. when the user tries to access them and they are not defined as innate features). Internally, dclab instantiates an RTDCBase subclass as defined by the format key. For mapped basins, dclab additionally creates a hierarchy child from the original dataset by filling the manual filtering array with the mapping information. To see which features are defined in basins, you can check the RTDCBase.features_basin property. The basins are directly accessible via RTDCBase.basins (and the basin datasets via RTDCBase.basins[index].ds).