marcenacp's picture
Deploy (see actual commits on https://github.com/mlcommons/croissant).
bc133ae
raw
history blame
12.8 kB
import enum
from typing import Any
import streamlit as st
from core.state import Field
from core.state import RecordSet
from events.fields import ExtractType
from events.fields import FieldEvent
from events.fields import handle_field_change
from events.fields import TransformType
import mlcroissant as mlc
from utils import needed_field
_JSON_PATH_DOCUMENTATION = (
"The JSON path if the data source is a JSON (see"
" [documentation](https://www.ietf.org/archive/id/draft-goessner-dispatch-jsonpath-00.html))."
)
_EXTRACT_DOCUMENTATION = (
"The extraction method to get the value of the field (column in a CSV, etc)."
)
_COLUMN_NAME_DOCUMENTATION = "The name of the column if the data source is a CSV."
class SourceType:
"""The type of the source (distribution or field)."""
DISTRIBUTION = "distribution"
FIELD = "field"
EXTRACT_TYPES = [
ExtractType.COLUMN,
ExtractType.JSON_PATH,
ExtractType.FILE_CONTENT,
ExtractType.FILE_NAME,
ExtractType.FILE_PATH,
ExtractType.FILE_FULLPATH,
ExtractType.FILE_LINES,
ExtractType.FILE_LINE_NUMBERS,
]
TRANSFORM_TYPES = [
TransformType.FORMAT,
TransformType.JSON_PATH,
TransformType.REGEX,
TransformType.REPLACE,
TransformType.SEPARATOR,
]
def _get_extract(source: mlc.Source) -> str | None:
if source.extract.column:
return ExtractType.COLUMN
elif source.extract.file_property:
file_property = source.extract.file_property
if file_property == mlc.FileProperty.content:
return ExtractType.FILE_CONTENT
elif file_property == mlc.FileProperty.filename:
return ExtractType.FILE_NAME
elif file_property == mlc.FileProperty.filepath:
return ExtractType.FILE_PATH
elif file_property == mlc.FileProperty.fullpath:
return ExtractType.FILE_FULLPATH
elif file_property == mlc.FileProperty.lines:
return ExtractType.FILE_LINES
elif file_property == mlc.FileProperty.lineNumbers:
return ExtractType.FILE_LINE_NUMBERS
else:
return None
elif source.extract.json_path:
return ExtractType.JSON_PATH
return None
def _get_extract_index(source: mlc.Source) -> int | None:
extract = _get_extract(source)
if extract in EXTRACT_TYPES:
return EXTRACT_TYPES.index(extract)
return None
def _get_transforms(source: mlc.Source) -> list[str]:
transforms = source.transforms
return [_get_transform(transform) for transform in transforms]
def _get_transform(transform: mlc.Transform) -> str | None:
if transform.format:
return TransformType.FORMAT
elif transform.json_path:
return TransformType.JSON_PATH
elif transform.regex:
return TransformType.REGEX
elif transform.replace:
return TransformType.REPLACE
elif transform.separator:
return TransformType.SEPARATOR
return None
def _get_transforms_indices(source: mlc.Source) -> list[int]:
transforms = _get_transforms(source)
return [
TRANSFORM_TYPES.index(transform) if transform in TRANSFORM_TYPES else None
for transform in transforms
]
def _handle_remove_reference(field):
"""Removes the reference from a field."""
field.references = mlc.Source()
def render_source(
record_set: RecordSet,
field: Field,
possible_sources: list[str],
):
"""Renders the form for the source."""
source = field.source
prefix = f"source-{record_set.name}-{field.name}"
col1, col2, col3 = st.columns([1, 1, 1])
index = (
possible_sources.index(source.uid) if source.uid in possible_sources else None
)
options = [s for s in possible_sources if not s.startswith(record_set.name)]
if index and (index < 0 or index >= len(options)):
index = None
key = f"{prefix}-source"
col1.selectbox(
needed_field("Data source"),
index=index,
options=options,
key=key,
help=(
"Data sources can be other resources (FileObject, FileSet) or other fields."
),
on_change=handle_field_change,
args=(FieldEvent.SOURCE, field, key),
)
if source.node_type == "distribution":
extract = col2.selectbox(
needed_field("Extract"),
index=_get_extract_index(source),
key=f"{prefix}-extract",
help=_EXTRACT_DOCUMENTATION,
options=EXTRACT_TYPES,
on_change=handle_field_change,
args=(FieldEvent.SOURCE_EXTRACT, field, key),
)
if extract == ExtractType.COLUMN:
key = f"{prefix}-columnname"
col3.text_input(
needed_field("Column name"),
value=source.extract.column,
key=key,
help=_COLUMN_NAME_DOCUMENTATION,
on_change=handle_field_change,
args=(FieldEvent.SOURCE_EXTRACT_COLUMN, field, key),
)
if extract == ExtractType.JSON_PATH:
key = f"{prefix}-jsonpath"
col3.text_input(
needed_field("JSON path"),
value=source.extract.json_path,
key=key,
help=_JSON_PATH_DOCUMENTATION,
on_change=handle_field_change,
args=(FieldEvent.SOURCE_EXTRACT_JSON_PATH, field, key),
)
# Transforms
indices = _get_transforms_indices(field.source)
if source.transforms:
for number, (index, transform) in enumerate(zip(indices, source.transforms)):
_, col2, col3, col4 = st.columns([4.5, 4, 4, 1])
key = f"{prefix}-{number}-transform"
selected = col2.selectbox(
"Transform",
index=index,
key=key,
options=TRANSFORM_TYPES,
on_change=handle_field_change,
help="One or more transformations to apply after extracting the field.",
args=(FieldEvent.TRANSFORM, field, key),
kwargs={"number": number},
)
if selected == TransformType.FORMAT:
key = f"{prefix}-{number}-transform-format"
col3.text_input(
needed_field("Format a date"),
value=transform.format,
key=key,
on_change=handle_field_change,
help=(
"For dates, use [`Python format"
" codes`](https://docs.python.org/3/library/datetime.html#strftime-and-strptime-format-codes)."
),
args=(selected, field, key),
kwargs={"number": number},
)
elif selected == TransformType.JSON_PATH:
key = f"{prefix}-{number}-jsonpath"
col3.text_input(
needed_field("JSON path"),
value=transform.json_path,
key=key,
on_change=handle_field_change,
help=_JSON_PATH_DOCUMENTATION,
args=(selected, field, key),
kwargs={"number": number},
)
elif selected == TransformType.REGEX:
key = f"{prefix}-{number}-regex"
col3.text_input(
needed_field("Regular expression"),
value=transform.regex,
key=key,
on_change=handle_field_change,
help=(
"A regular expression following [`re` Python"
" convention](https://docs.python.org/3/library/re.html#regular-expression-syntax)"
" with one capturing group. The result of the operation will be"
" the last captured group."
),
args=(selected, field, key),
kwargs={"number": number},
)
elif selected == TransformType.REPLACE:
key = f"{prefix}-{number}-replace"
col3.text_input(
needed_field("Replace pattern"),
value=transform.replace,
key=key,
on_change=handle_field_change,
help=(
"A replace pattern separated by a `/`, i.e."
" `string_to_replace/string_to_substitute` in order to replace"
" `string_to_replace` by `string_to_substitute`."
),
args=(selected, field, key),
kwargs={"number": number},
)
elif selected == TransformType.SEPARATOR:
key = f"{prefix}-{number}-separator"
col3.text_input(
needed_field("Separator"),
value=transform.separator,
key=key,
on_change=handle_field_change,
help="A separator to split strings on, e.g. `|` to split `a|b|c`.",
args=(selected, field, key),
kwargs={"number": number},
)
def _handle_remove_transform(field, number):
del field.source.transforms[number]
col4.button(
"✖️",
key=f"{prefix}-{number}-remove-transform",
help="Remove the transformation.",
on_click=_handle_remove_transform,
args=(field, number),
)
def _handle_add_transform(field):
if not field.source:
field.source = mlc.Source(transforms=[])
field.source.transforms.append(mlc.Transform())
col1, _, _ = st.columns([1, 1, 1])
col1.button(
"Add transform on data",
key=f"{prefix}-close-fields",
help="Add a transformation.",
on_click=_handle_add_transform,
args=(field,),
)
def render_references(
record_set: RecordSet,
field: Field,
possible_sources: list[str],
):
"""Renders the form for references."""
key = f"references-{record_set.name}-{field.name}"
button_key = f"{key}-add-reference"
has_clicked_button = st.session_state.get(button_key)
references = field.references
if references or has_clicked_button:
col1, col2, col3, col4 = st.columns([4.5, 4, 4, 1])
index = (
possible_sources.index(references.uid)
if references.uid in possible_sources
else None
)
options = [s for s in possible_sources if not s.startswith(record_set.name)]
if index and (index < 0 or index >= len(options)):
index = None
key = f"{key}-reference"
col1.selectbox(
"Reference",
index=index,
options=options,
key=key,
on_change=handle_field_change,
args=(FieldEvent.REFERENCE, field, key),
)
if references.node_type == "distribution":
key = f"{key}-extract-references"
extract = col2.selectbox(
needed_field("Extract the reference"),
index=_get_extract_index(references),
key=key,
options=EXTRACT_TYPES,
help=_EXTRACT_DOCUMENTATION,
on_change=handle_field_change,
args=(FieldEvent.REFERENCE_EXTRACT, field, key),
)
if extract == ExtractType.COLUMN:
key = f"{key}-columnname"
col3.text_input(
needed_field("Column name"),
value=references.extract.column,
key=key,
help=_COLUMN_NAME_DOCUMENTATION,
on_change=handle_field_change,
args=(FieldEvent.REFERENCE_EXTRACT_COLUMN, field, key),
)
if extract == ExtractType.JSON_PATH:
key = f"{key}-jsonpath"
col3.text_input(
needed_field("JSON path"),
value=references.extract.json_path,
key=key,
help=_JSON_PATH_DOCUMENTATION,
on_change=handle_field_change,
args=(FieldEvent.REFERENCE_EXTRACT_JSON_PATH, field, key),
)
col4.button(
"✖️",
key=f"{key}-remove-reference",
help="Remove the join.",
on_click=_handle_remove_reference,
args=(field,),
)
elif not has_clicked_button:
st.button(
"Add a join with another column/field",
key=button_key,
)