Storing and sharing data is obviously rather important. Hence we need a file-format which enables us to easily read, write and share the data present in _tacoma.edge_lists and _tacoma.edge_changes.

## The status quo — csv¶

Typically, data is shared in csv-format (or tsv) where the first column refers to a discrete time, the second column contains a source node and the third column contains a target node, such as

A csv-example for edge lists
time source target
20 Alice Bob
20 Alice Clara
60 Darren Alice

These formats usually describe edge lists at discrete times.

For data sets describing edge changes, this format often includes a fourth column containing a descriptor for whether the edge was activated or deactivated at the corresponding time in the first columns, such as

A csv-example for edge changes
time source target in or out
20 Alice Bob in
20 Alice Clara in
60 Alice Bob out
60 Alice Clara out
60 Darren Alice in

The only advantage of these formats is that they’re easily readable by humans.

• do not contain the total number of nodes
• no description of the data, information about the experiment
• ambiguous dimension of time
• final time of the experiment $$t_\mathrm{max}$$ is not given
• ambiguous about the class of the provided data

## A new format — The taco¶

The number of cases in which data is opened to be read by a human is significantly lower than the number of cases in which it is opened to be read by a machine. My personal opinion is therefore that data should be stored in a format easily readible by a computer and converted to appropriate formats whenever a human wants to access the data directly.

tacoma comes with its own data-format called “.taco” which is simply the whole data of either a _tacoma.edge_lists or a _tacoma.edge_changes object dumped to a JSON object and then written to a file. We chose JSON since there exists packages to read and write JSON in nearly every programming environment. Furthermore, JSON-strings are considerably more light-weight than, e.g. XML-strings.

Hence, not unlike to a real taco, we dump all the good stuff to a single shell — delicious, but difficult for human consumption.

### Writing¶

In tacoma, we can save a temporal network to a taco using tacoma.data_io.write_json_taco().

tc.write_json_taco(temporal_network, '/path/to/temporal_network.taco')


The fields in the written JSON object are the same as they are in the formats defined in “Temporal network classes”, besides the additional field 'type' which can either be 'edge_lists' or 'edge_changes'.

Considering the example temporal network given in “Temporal network classes” the JSON-string for _tacoma.edge_lists is congruent to

{
'type': 'edge_lists',
't': [0.0, 1.0, 1.5, 3.0, 4.0, 7.0, 7.31],
'tmax': 8.1,
'N': 8,
'edges': [ [[0, 1], [1, 7]],
[[0, 1]],
[[0, 1], [1, 7]],
[[2, 5], [1, 7]],
[[2, 5]],
[[0, 1], [2, 5]],
[[0, 1]]
],
'int_to_node': {
'0': 'Alice',
'1': 'Bob',
'2': 'Clara',
'3': 'Darren',
'4': 'Elle',
'5': 'Felicitas',
'6': 'George',
'7': 'Harriett'
},
'notes': 'This experiment was conducted as a test.',
'time_unit': 's'
}


The actual data is, however, minified and looks more like

{"type":"edge_lists","t":[0.0,1.0,1.5,3.0,4.0,7.0,7.31],"tmax":8.1,"N":8,"edges":[[[0,1],[1,7]],[[0,1]],[[0,1],[1,7]],[[2,5],[1,7]],[[2,5]],[[0,1],[2,5]],[[0,1]]],"int_to_node":{"0":"Alice","1":"Bob","2":"Clara","3":"Darren","4":"Elle","5":"Felicitas","6":"George","7":"Harriett"},"notes":"This experiment was conducted as a test.","time_unit":"s"}


For _tacoma.edge_changes it would look like

{
'type': 'edge_changes',
't': [1.0, 1.5, 3.0, 4.0, 7.0, 7.31],
't0': 0.0,
'tmax': 8.1,
'N': 8,
'edges_initial': [[0, 1], [1, 7]],
'edges_in': [[], [[1, 7]], [[2, 5]], [], [[0, 1]], []],
'edges_out': [[[1, 7]], [], [[0, 1]], [[1, 7]], [], [[2, 5]]],
'int_to_node': {
'0': 'Alice',
'1': 'Bob',
'2': 'Clara',
'3': 'Darren',
'4': 'Elle',
'5': 'Felicitas',
'6': 'George',
'7': 'Harriett'
},
'notes': 'This experiment was conducted as a test.',
'time_unit': 's'
}


The actual data is, however, minified and looks more like

{"type":"edge_changes","t":[1.0,1.5,3.0,4.0,7.0,7.31],"t0":0.0,"tmax":8.1,"N":8,"edges_initial":[[0,1],[1,7]],"edges_in":[[],[[1,7]],[[2,5]],[],[[0,1]],[]],"edges_out":[[[1,7]],[],[[0,1]],[[1,7]],[],[[2,5]]],"int_to_node":{"0":"Alice","1":"Bob","2":"Clara","3":"Darren","4":"Elle","5":"Felicitas","6":"George","7":"Harriett"},"notes":"This experiment was conducted as a test.","time_unit":"s"}


Reading temporal network data from a taco is as simple as using tacoma.data_io.load_json_taco().

temporal_network = tc.load_json_taco('temporal_network.taco')


### Converting csv to taco¶

As indicated above, converting csv-data to data actually usable by algorithms can turn out quite tideous. Below you can find a commented example on how to load a csv-file (here from the SocioPatterns ‘Hypertext 2009’-dataset) and convert it to a taco, taken directly from tacoma.data_io

import gzip
import csv

import tacoma as tc

# open gzipped file
gzip_file = 'ht09_contact_list.dat.gz'
with gzip.open(gzip_file,mode='rt') as f:

# mappings of nodes to integers
node_to_int = {}
int_to_node = {}

# get an initial t_old
# (this is done to detect changes in the tsv)
t_old = None

# list of edge lists
edges = []

# time points
time = []
t = float( int(row[0]) - 20 ) #this is to account for the interval choice [t-20s, t]

# if the time changed, we save the new time and
# prepare to save new edges
if t_old != t:

# When the time changed more than dt,
# append an instance of an empty edge list
# at t = t_old + dt
if (t_old is not None) and (t - t_old > 20):
edges.append([])
time.append(t_old+20)

edges.append([])
time.append(t)

# get the edge
i = int(row[1])
j = int(row[2])

# map the edge to integers
if i not in node_to_int:
this_int = len(node_to_int)
node_to_int[i] = len(node_to_int)
int_to_node[this_int] = str(i)

if j not in node_to_int:
this_int = len(node_to_int)
node_to_int[j] = len(node_to_int)
int_to_node[this_int] = str(j)

# save the edge
edges[-1].append(tuple(sorted([
node_to_int[i],
node_to_int[j]
])))
t_old = t

N = len(node_to_int)
tmax = time[-1] + 20.0

# get a new edge_lists instance
el = tc.edge_lists()

el.N = N
el.tmax = tmax
el.edges = edges
el.t = time
el.time_unit = 's'
el.notes = """
This data is binned.

In this data, t0 = 0.0 corresponds to 8am on Jun 29th 2009 (UNIX time 1246255200).

If you use this data, please cite

L. Isella et al.,  What's in a crowd? Analysis of face-to-face behavioral networks,
Journal of Theoretical Biology 271, 166 (2011).
"""
el.int_to_node = int_to_node

# verifying that this is a valid temporal network
tc.verify(el)

# save this edge_lists instance
with open('ht09.taco','w') as f:
tc.write_json_taco(el,f)


## Edge coordinate files¶

tacoma.data_io.write_edge_trajectory_coordinates()