Before diving into the detail of this doc, you're strongly recommended to know some important concepts about system analyses.
In this file we describe how to analyze text classification models.
We will give an example using the text-classification
sst2 dataset,
but other datasets can be analyzed in a similar way.
-
(1)
datalab
: if your datasets have been supported by datalab, you fortunately don't need to prepare the dataset. -
(2)
tsv
(without column names at the first row), see one example
I love this movie positive
The movie is too long negative
...
- (3)
json
(basically, it's a list of dictionaries with two keys:text
andtrue_label
)
[
{"text": "I love this movie", "true_label": "positive"},
{"text": "The movie is too long", "true_label": "negative"}
...
]
In this task, your system outputs should be as follows:
- (1)
text
: one predicted label per line
predicted_label
- (2)
json
: a list of dictionaries with one key:predicted_label
)
[
{"predicted_label": "positive"},
{"predicted_label": "negative"}
...
]
Let's say we have several files such as
etc. from different systems.
The below example loads the sst2
dataset from DataLab:
explainaboard --task text-classification --dataset sst2 --system-outputs ./data/system_outputs/sst2/sst2-lstm-output.txt
where
--task
: denotes the task name, you can find all supported task names here--system-outputs
: denote the path of system outputs. Multiple one should be separated by space, for example, system1 system2--dataset
: denotes the dataset namereport.json
: the generated analysis file with json format. Tips: use a json viewer like this one for better interpretation.
Alternatively, you can load the dataset from an existing file using the
--custom-dataset-paths
option
explainaboard --task text-classification --custom-dataset-paths ./data/system_outputs/sst2/sst2-dataset.tsv --system-outputs ./data/system_outputs/sst2/sst2-lstm-output.txt
in which case the file format of this file is TSV
text \t true_label
One also can perform pair-wise analysis:
explainaboard --task text-classification --dataset sst2 --system-outputs ./data/system_outputs/sst2/sst2-lstm-output.txt ./data/system_outputs/sst2/sst2-cnn-output.txt > report.json
where two system outputs are fed separated by space.
report.json
: the generated analysis file with json format, whose schema is similar to the above one with single system evaluation except that all performance values are obtained using the sys1 subtract sys2.