Skip to main content

WEAVE-IO query interface for accessing WEAVE data

Project description

PyPI

Installation

Installation is done using pip with Python 3.7 or above (python environment installation is handled in the install script). The install script install.sh can be downloaded and run with chmod +x install.sh && ./install.sh. This will place weaveio in its own environment - accessed using conda activate weaveio. Upgrades are now handled by the weaveio command line application which will be installed along with weaveio.

To enable ssh access for jupyter notebooks please refer to the section entitled "ssh access" below.

Upgrading to 2022.1.5

If you are on a weaveio version older than 2022.1.5 then please run the following:

  1. conda activate weaveio
  2. pip install weaveio --upgrade --no-cache-dir
  3. Remove old weaveio aliases in your .bashrc or .tcshrc files
  4. Add a new alias with conda activate weaveio && echo "alias weaveio=$(dirname "$(which python)")/weaveio" >> ~/.bashrc
  5. source ~/.bashrc or source ~/.tcshrc

weaveio command line

In versions post 2022.1.5, we will use the weaveio command line to run the following commands:

  • console - runs an ipython console with weaveio activated
  • jupyter - starts a jupyter notebook in the current directory in the weaveio environment (you can list active servers and also stop them).
  • version - Returns version info
  • upgrade - Uses pip to install the latest version of weaveio

Tutorial

WEAVE objects

  • An OB holds all the information that pertains to making an observation: the targets, the conditions, the instrument configuration. You can locate specific OBs with their obid data.obs[obid]

  • An Exposure is one integration of both arms of the spectrograph. You can locate Exposures like so: data.exposures[mjd]

  • A Run is a weave term for an exposure taken in one arm of the spectrograph (so there are always 2 Runs per Exposure). You can locate runs using their runid data.runs[runid].

  • A spectrum in weaveio refers to a single spectrum of a single target (not the block of spectr*a*)

  • An L1 product refers to the spectra recorded in an L1File. All L1 data is separated by camera colour (red and blue).

    • A single spectrum is the processed spectrum from the raw data
    • A stack spectrum is the spectrum resulting from stacking two or more single spectra in a single ob
    • A superstack spectrum results from stacking between OBs but with the same instrument configuration
    • A supertarget spectrum results from stacking every single spectrum of a single Weave_Target cname.
  • An L2 product refers to analysis products performed by Redrock, Gandalf, PPXF, RVSpecFit, and Ferre.

  • Each one of these fits has a model spectrum and analysis output (such as line indices)

  • Each L2 product has at least 2 corresponding L1 products since the red and blue spectra are joined together for the analysis.

  • There are three types of target

    1. weave_target is the unified target based on ra, dec. They have a unique CNAME
    2. survey_target is a target specified by a survey in a survey_catalogue (they reference a single weave_target). These are unique to a catalogue.
    3. fibre_target is a result of assigning a spectrograph fibre to a survey_target. These are unique to an OBSpec.

What is an attribute? What is an object? What is a product?

  • weave.io is an object orientated query language using a neo4j database hosted at lofar.herts.ac.uk.
  • weave.io stores 4 (+1) types of object:
    1. File - A reference to a physical fits file on the herts system (it also stores references to individual fits.HDUs and their headers as separate objects, but the user doesn't need to know of their existence to use weave.io)
    2. Object - A object that references a concept in the WEAVE universe that has attributes (an OB object has an obid attribute, an exposure object has an attribute expmjd
    3. Attribute - A piece of data that belongs to some object
    4. Product - A special type of attribute which references binary data not stored in the database itself (e.g. spectrum flux). You cannot perform arithmetic/indexing on product attributes.

Running a query

A query finds the locations of all the L1/L2/Raw products that you want. It is analogous to an SQL query except that it is written in Python.

  • A query is constructed using python like so:

    from weaveio import *
    data = Data(username, password)
    
    runs = data.obs[obid].runs
    reds = runs[runs.colour == 'red']
    spectra = reds.l1singlespectra
    

    runs, reds, spectra are all queries

  • Each part of this query can be run independently, using the parentheses:

    • runs.runids is still a query
    • runs.runids() returns an actual list of numbers
    • reds() will return a list of Run objects (containing all attributes of a run)

Examples of use:

1. I want to return the number of sky spectra in a given run (runid=1002850)

from weaveio import *
data = Data() 
runid = 1003453
nsky = sum(data.runs[runid].targuses == 'S')
print("number of sky targets = {}".format(nsky()))

output: number of sky targets = 100

We can break this down into several steps:

  1. from weaveio import *; data = Data() - Import all the necessary weaveio functions and start the default lofar database link (the default is opr3 but this may change in the future).
  2. data. - Start building a query using data connection established above
  3. data.runs - Get all runs
  4. data.runs[runid] - Filter the runs to those that have their id equal to the variable runid. Each run has a unique runid, so you can be sure that this query now contains one row.
  5. data.runs[runid].targuses - Each run has multiple L1 single spectra associated with it and each of those spectra have a targuse attribute. Therefore, each run has multiple targuse attributes, therefore you must write targuses.
  6. data.runs[runid].targuses == 'S' - Make a boolean mask for where the targuse flag for each spectrum belonging to this run is set to 'S' (this refers to "sky").
  7. nsky = sum(data.runs[runid].targuses == 'S') - Sum the entire boolean mask, thereby counting the number of sky fibres placed in this run. The python function sum was overwritten with a weaveio version when we did our imports. sum is now compatible with weaveio but can also be used normally.
  8. nsky() - Up til now, we have been building a query, much like we would write SQL, but nothing has executed on the database yet. To run our query and fetch the result, we call it using the parentheses ().

1b. I want to see how many sky targets each run has

from weaveio import *
data = Data()
nsky = sum(data.runs.targuses == 'S', wrt=data.runs)  # sum the number of sky targets with respect to their runs
print(nsky())

output: [100 299 299 100 100 200 160 ...]

This query is very similar to the previous one except that we are summing over the fibres of each run, not just 1 run as before. The difference here is that we have missed out data.runs[runid] which means that our query references all runs in the database at once.

  1. from weaveio import *; data = Data() - Import all the necessary weaveio functions and start the default lofar database link.
  2. data.runs - Get all runs.
  3. data.runs.targuses == 'S - Access all targuse attributes belonging to each run. Read this statement as "for each run in data, for each targuse in run, do ==S.
  4. nsky = sum(data.runs.targuses == 'S', wrt=data.runs) - This time sum our boolean mask with respect to (wrt) data.runs. This means each row in the resultant query, nsky, will refer to each row in data.runs. I.E. There is now a query row per run, whereas in the previous example there was only one row.

1c. Put the above result into a table where I can see the runid

from weaveio import *
data = Data()
nsky = sum(data.runs.targuses == 'S', wrt=data.runs)  # sum the number of skytargets with respect to their runs
query_table = data.runs[['id', nsky]]  # design a table by using the square brackets
concrete_table = query_table()  # make it "real" by executing the query
print(concrete_table)
print(type(concrete_table))

output:

   id   sum0
------- ----
1003453  100
1003440  299
...      ...
<class 'weaveio.readquery.results.Table'>  # although this is an astropy table really

Returning more than one attribute per row requires "designing" a table. To do this, we put a list of our required values in the square brackets [['id', nsky]]. Any string referring to an attribute (e.g. 'id') can go here as well as any previously written query (e.g. nsky'). However, any items that you put in the square brackets must align with the object outside:

For example:

  • data.runs[['id', nsky]] is valid because each run has an id and the query nsky is based on data.runs (i.e. each run has an nsky calculated for it).

2. I want to plot all single sky spectra from last night in the red arm

from weaveio import *
data = Data()
yesterday = 57811  # state yesterday's date in MJD

runs = data.runs
is_red = runs.camera == 'red'
is_yesterday = floor(runs.exposure.mjd) == yesterday  # round down to an integer, which is the day

runs = runs[is_red & is_yesterday]  # filter the runs to red ones that were taken yesterday  
spectra = runs.l1single_spectra  # get all the spectra per run
sky_spectra = spectra[spectra.targuse == 'S']  # filter to the spectra which are sky 

table = sky_spectra[['wvl', 'flux']]  # design a table of wavelength and flux

import matplotlib.pyplot as plt
# this may take a while to plot, there is a lot of data
for row in table:  # you can iterate over a query with `for` as well as requesting the whole thing with `()` 
    plt.plot(row.wvl, row.flux, 'k-', alpha=0.4)  # standard matplotlib line plot 
plt.savefig('sky_spectra.png')

output:

The only new thing in this query is for row in table. This implicitly calls the table (table()) and downloads one row at a time. You will want to do this when the resulting query will be large. By using this "iterator" pattern, you can avoid loading it all into memory at once.

3. I want to plot the H-alpha flux vs. L2 redshift distribution from all WL or W-QSO spectra that were observed from all OBs observed in the past month. Use the stacked data

import matplotlib.pyplot as plt
data = Data()
l2s = data.l2stacks
l2s = l2s[(l2s.ob.mjd >= 57780) & any(l2s.fibre_target.surveys == '/WL.*/', wrt=l2s.fibre_target)]
l2s = l2s[l2s['ha_6562.80_flux'] > 0]
table = l2s[['ha_6562.80_flux', 'z']]()
plt.scatter(table['z'], table['ha_6562.80_flux'], s=1)
plt.yscale('log')
plt.savefig('ha-z.png')

Let's break down this query:

  1. l2s = data.l2stacks gets all l2stack products in the database. These are the data products which contain joined spectra and template fits.
  2. l2s.fibre_target.surveys == '/WL.*/' - This creates a boolean mask matching the survey name to 'WL.*' with regex. You can activate regex by using / at the start and end of a string.
  3. l2s = l2s[(l2s.ob.mjd >= 57780) & any(l2s.fibre_target.surveys == '/WL.*/', wrt=l2s.fibre_target)] - This filters to l2 products whose L1 observations were taken after 57780 and survey names containing "WL"
  4. l2s = l2s[l2s['ha_6562.80_flux'] > 0] - Then we further filter the l2 products by required an halpha flux greater than 0 (fit by Gandalf).
  5. l2s[['ha_6562.80_flux', 'z']] - This designs a table with the halpha flux (from gandalf) and the redshift (from redrock)

4a. Join on a 3rd party catalogue

Given a catalogue of weave cnames, find those objects in the database and return the calendar dates on which those matched objects were observed, and the number of WEAVE visits to each CNAME (there could be more than one)

To do this we need to use join which is imported from weaveio. join takes at least 3 arguments: the first is the table to join on, the second is the column name in that table, and the third is the object in weaveio to join to. You may also specify a join_query which is another weaveio query that results in the attribute to join to. If this is not specified, then it is assumed that the attribute should be the same as the column name in the table.

def join(table: Table, index_column: str,
         object_query: ObjectQuery, join_query: Union[AttributeQuery, str] = None,
         join_type: str = 'left') -> Tuple[TableVariableQuery, ObjectQuery]:
    ...

The output of join is the input table converted to a weaveio variable and a reduced version of the input object_query. The output table variable should now be treated as rows.

from astropy.table import Table
from weaveio import *
import weaveio
fname = Path(weaveio.__file__).parents[0] / 'tests/my_table.ascii'
data = Data()
table = Table.read(fname, format='ascii')
rows, targets = join(table, 'cname', data.weave_targets)
mjds = targets.exposures.mjd  # get the mjd of the plate exposures for each target
q = targets['cname', rows['modelMag_i'], {'mjds': mjds, 'nobservations': count(mjds, wrt=targets)}]
print(q())

output:

       cname         modelMag_i          mjds [15]           nobservations
-------------------- ---------- ---------------------------- -------------
WVE_10461805+5755400   20.20535 57809.109711 .. 57811.075961            15
WVE_10521675+5814292    21.2665 57809.109711 .. 57811.075961            15
WVE_10521675+5814292    21.2665 57809.109711 .. 57811.075961            15
WVE_02175674-0451074   21.38155             57640.1764 .. --             6
WVE_02174727-0459587   21.81214             57640.1764 .. --             6
WVE_02175411-0504122   22.28189             57640.1764 .. --             6
WVE_02175687-0512209   21.79577             57640.1764 .. --             6
WVE_02174991-0454427   21.65417             57640.1764 .. --             6
WVE_02175370-0448267   19.63735             57640.1764 .. --             6
WVE_02174862-0457336     22.181             57640.1764 .. --             6
WVE_02175320-0508011   20.16733             57640.1764 .. --             6

Breaking down this query:

  1. table = Table.read('weaveio/tests/my_table.ascii', format='ascii') - This reads in a custom table from the file my_table.ascii. One of the column names is cname.
  2. rows, targets = join(table, 'cname', data.weave_targets) - This joins the cname column of the table to the cname attribute of the weave targets catalogue. targets will refer to all weave_targets that were matched by the cname column and rows will refer to the rows of the table.
  3. mjds = targets.exposures.mjd - This gets the mjd of the plate exposures for each target (there may be more than 1) and each exposure will have 2 l1single_spectra (one for each arm), although we don't worry about that yet.
  4. q = targets['cname', rows['modelMag_i'], {'mjds': mjds, 'nobservations': count(mjds, wrt=targets)}] - This creates a table using the 'modelMag_i' found in the fits file table. This can be done because we joined it earlier. Here we are also renaming columns to more human readable names using a dictionary.

Ragged arrays

The mjd result column is "ragged" array since there may be more than 1 exposure per target and that is not constant for each target. So that the user can aggregate easily we convert the mjd result column to a regular array and mask it.

4b. Plot sdss modelMag_i from the fits file against mean flux between 400-450nm

Continuing from 4a, we first traverse to the l1single_spectra and fetch their wavelengths and fluxes. Then we plot the modelMag_i from the fits file against the mean flux between 400-450nm.

import matplotlib.pyplot as plt

q = targets.l1single_spectra[['cname', rows['modelMag_g'], 'wvl', 'flux', 'sensfunc']]
table = q()
mean_fluxes = []
for row in table:
    filt = (row['wvl'] > 4000) & (row['wvl'] < 4500)  # angstroms
    mean_fluxes.append(mean(row['flux'][filt]))
table['mean_flux'] = mean_fluxes
print(table['mean_flux'])
plt.scatter(table['modelMag_g'], -2.5 * np.log10(table['mean_flux']))
plt.show()

output:

      mean_flux      
---------------------
   1.6613570457103553
   1.8225295509082666
   1.6668027617324288
                   --
   1.8113559953805027
                   --
    1.685038564203977
                  ...
                   --
 -0.07946323931008473
                   --
-0.012973852190988072
 -0.13294200506014725
                   --
                   --
Length = 90 rows

5. For each OB at a time, retrieve all the stacked red-arm sky spectra and the single spectra that went into making those stacked spectra

from weaveio import *
data = Data()

obs = split(data.obs)  # mark the fact that you want have one table per OB thereby "splitting" the query in to multiple queries
stacks = obs.l1stack_spectra[(obs.l1stack_spectra.targuse == 'S') & (obs.l1stack_spectra.camera == 'red')]
singles = stacks.l1single_spectra
singles_table =  singles[['flux', 'ivar']]
query = stacks[['ob.id', {'stack_flux': 'flux', 'stack_ivar': 'ivar'}, 'wvl', {'single_': singles_table}]]

for index, ob_query in query:
    print(f"stacks and singles for OB #{index}:")
    print(ob_query())

output:

stacks and singles for OB #3133:
ob.id stack_flux [15289] ... single_flux [3,15289] single_ivar [3,15289]
----- ------------------ ... --------------------- ---------------------
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0
 3133         0.0 .. 0.0 ...            0.0 .. 0.0            0.0 .. 0.0

Breaking down this query:

There are two new concepts in this example: query splitting and adding tables together.

  1. Splitting occurs with obs = split(data.obs). Nothing special happens here except that we have now marked that any query that follows from obs will yield more than one table. Each table will have a different ob.id value
  2. We continue our query as normal
  3. query = stacks[['ob.id', {'stack_flux': 'flux', 'stack_ivar': 'ivar'}, 'wvl', {'single_': singles_table}]] - Here we have now added the singles_table into a new table we are constructing. This is equivalent to query = stacks[['ob.id', {'stack_flux': 'flux', 'stack_ivar': 'ivar'}, 'wvl', {'single_flux': singles['flux'], 'single_ivar': singles['ivar']}]]. When renaming the additional table (with {'single_': ...}) we are added a prefix onto each of the new columns.
  4. for index, ob_query in query: - query is now split query, so when we iterate over it we get one table for each ob.id value. It also returns an index, which in this case is just the ob.id value. ob_query is now identical to the original query except that is will only return results for one OB.
  5. ob_query() - Execute the query only for the current OB (the one with ob.id == index).

split is the equivalent function to group_by in pandas or astropy. However, you must perform a split before querying whereas in a pandas/astropy group_by it is done after the fact.

Also, it is important to note that the single_flux and single_ivar columns are 2 dimensional since there are 3 single spectra per stack spectrum. So you get all 3 at once, per row of the query.

Details

If confused, ignore...

object/attribute

weave.io uses Python syntax to traverse a hierarchy of objects and their attributes. It is important to note that objects are linked to other objects (e.g. a run belongs to an OB and also to an exposure, which itself belongs to an OB).

You can stitch together these objects to form a hierarchy of objects:

run <-- exposure <-- ob <--obspec(xml)

Every OB is a parent of multiple Exposures which in turn are parents exactly 2 runs each (one red, one blue).

Because of this chain of parentage/relation, every object has access to all attributes where there is a chain, as if they were its own attributes.

Traversal syntax

  1. You can request any directly owned attribute of an object

    • An OB has an id: ob.id
    • An obstemp has a maxseeing obstemp.maxseeing
  2. You can request any attribute of objects that are further away in the hierarchy as if it were its own. This is useful because a priori you wouldn't be expected to know where any particular piece of data is stored, just that it exists.

    • run.maxseeing is identical to run.exposure.ob.obspec.obstemp.maxseeing
  3. Traversal works in any direction

    • You can go down a hierarchy: exposure.runs.raw_spectrum (exposure has multiple runs)
    • You can go up as well: raw_spectrum.run.exposure (raw_spectrum has one run)
  4. Traversal can be implicit like with the indirectly accessed attributes above

    • You can skip out stages: run.obspec is identical to run.ob.obspec
  5. Traversal requires you to be aware of plurality/multiplicity (neo4j calls this cardinality):

    • A run only ever has a single ob, so you query it using a singular name: run.ob
    • But an ob will typically have multiple runs, so you must use plural names: ob.runs
    • weave.io is aware of plurality of the whole hierarchy, so it will shout at you if you are obviously wrong: ob.run will fail before you even execute the query.
  6. Traversal name plurality is relative

    • A run has a single ob, which in turn has multiple runs: run.ob.runs will return all runs of the ob (including the one that was explicitly referenced at the start).
    • ob.runs.weave_target.obs can be read as "For each of the runs, get its weave_target, and then for each weave_target get all OBs which assign a fibre to that target."
  7. Traversal using dot syntax always increases/maintains the total number of rows returned at the end

    • A consequence of traversal is the building up of rows. This is useful to get properly aligned labels/indexes for things.
    • ob.runs.ob.runs.ob.runs will return not simply return the runs of this ob, but rather a huge duplicated list because each time you use the '.' syntax, we are saying "for each"

Identifiers

  1. You can request a specific object if you know its id
    • one_ob = data.obs[obid]
    • a_list_of_obs = data.obs[[obid1, obid2, obid3]]
    • Plurality still applies here: data.weave_targets.obs[obid] will return one ob for each weave_target
      • data.obs[obid].weave_targets returns all weave_targets for this particular ob
      • data.weave_targets.obs[obid] returns the ob with obid for each weave_target (sometimes will be None)

SSH access

instructions to enable passwordless ssh access for jupyter weaveio:

  • open/create directories ~/.ssh/config
  • add the following to that file
		Host lofar
	        User <USERNAME>
	        HostName lofar.herts.ac.uk
	        LocalForward <PERSONALPORT> 127.0.0.1:<PERSONALPORT>
	        LocalForward 7474 127.0.0.1:7474
	        LocalForward 7687 127.0.0.1:7687
	        ForwardX11 yes
  • ls -al ~/.ssh
  • if no key files are listed there:
    • ssh-keygen -t ed25519 -C "<your_email@example.com>"
    • eval "$(ssh-agent -s)"
    • ssh-add ~/.ssh/id_ed25519
  • open public key file (~/.ssh/id_ed25519.pub) and copy all contents to clipboard
  • ssh lofar
  • enter username and password
  • bash
  • nano ~/.ssh/authorized_keys
  • paste in your public key contents and save
  • nano ~/.bashrc
    • append export DISPLAY=localhost:0.0
  • logout and login to verify

machine-readable-change-log ########################### 2022.1.7: Added align helper function; fixed problem with NoSS access; added better names to user-unnamed columns; tests

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

weaveio-2022.1.7.tar.gz (166.8 kB view hashes)

Uploaded source

Built Distribution

weaveio-2022.1.7-py2.py3-none-any.whl (180.9 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page