mirror of
https://github.com/gradio-app/gradio.git
synced 2024-11-27 01:40:20 +08:00
18cc8a385c
* Add assertion for file_types * Changelog * Add logic to UploadButton Co-authored-by: Abubakar Abid <abubakar@huggingface.co>
2351 lines
81 KiB
Python
2351 lines
81 KiB
Python
"""
|
|
Tests for all of the components defined in components.py. Tests are divided into two types:
|
|
1. test_component_functions() are unit tests that check essential functions of a component, the functions that are checked are documented in the docstring.
|
|
2. test_in_interface() are functional tests that check a component's functionalities inside an Interface. Please do not use Interface.launch() in this file, as it slow downs the tests.
|
|
"""
|
|
|
|
import filecmp
|
|
import json
|
|
import os
|
|
import pathlib
|
|
import shutil
|
|
import tempfile
|
|
from copy import deepcopy
|
|
from difflib import SequenceMatcher
|
|
from unittest.mock import MagicMock, patch
|
|
|
|
import matplotlib
|
|
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
import pandas as pd
|
|
import PIL
|
|
import pytest
|
|
import vega_datasets
|
|
from scipy.io import wavfile
|
|
|
|
import gradio as gr
|
|
from gradio import media_data, processing_utils
|
|
|
|
os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
|
|
matplotlib.use("Agg")
|
|
|
|
|
|
class TestComponent:
|
|
def test_component_functions(self):
|
|
"""
|
|
component
|
|
"""
|
|
assert isinstance(gr.components.component("textarea"), gr.templates.TextArea)
|
|
|
|
|
|
def test_raise_warnings():
|
|
for c_type, component in zip(
|
|
["inputs", "outputs"], [gr.inputs.Textbox, gr.outputs.Label]
|
|
):
|
|
with pytest.warns(UserWarning, match=f"Usage of gradio.{c_type}"):
|
|
component()
|
|
|
|
|
|
class TestTextbox:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, tokenize, generate_sample, get_config
|
|
"""
|
|
text_input = gr.Textbox()
|
|
assert text_input.preprocess("Hello World!") == "Hello World!"
|
|
assert text_input.postprocess("Hello World!") == "Hello World!"
|
|
assert text_input.postprocess(None) is None
|
|
assert text_input.postprocess("Ali") == "Ali"
|
|
assert text_input.postprocess(2) == "2"
|
|
assert text_input.postprocess(2.14) == "2.14"
|
|
assert text_input.serialize("Hello World!", True) == "Hello World!"
|
|
|
|
assert text_input.tokenize("Hello World! Gradio speaking.") == (
|
|
["Hello", "World!", "Gradio", "speaking."],
|
|
[
|
|
"World! Gradio speaking.",
|
|
"Hello Gradio speaking.",
|
|
"Hello World! speaking.",
|
|
"Hello World! Gradio",
|
|
],
|
|
None,
|
|
)
|
|
text_input.interpretation_replacement = "unknown"
|
|
assert text_input.tokenize("Hello World! Gradio speaking.") == (
|
|
["Hello", "World!", "Gradio", "speaking."],
|
|
[
|
|
"unknown World! Gradio speaking.",
|
|
"Hello unknown Gradio speaking.",
|
|
"Hello World! unknown speaking.",
|
|
"Hello World! Gradio unknown",
|
|
],
|
|
None,
|
|
)
|
|
assert text_input.get_config() == {
|
|
"lines": 1,
|
|
"max_lines": 20,
|
|
"placeholder": None,
|
|
"value": "",
|
|
"name": "textbox",
|
|
"show_label": True,
|
|
"type": "text",
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
assert isinstance(text_input.generate_sample(), str)
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface_as_input(self):
|
|
"""
|
|
Interface, process, interpret,
|
|
"""
|
|
iface = gr.Interface(lambda x: x[::-1], "textbox", "textbox")
|
|
assert iface("Hello") == "olleH"
|
|
iface = gr.Interface(
|
|
lambda sentence: max([len(word) for word in sentence.split()]),
|
|
gr.Textbox(),
|
|
"number",
|
|
interpretation="default",
|
|
)
|
|
scores = await iface.interpret(
|
|
["Return the length of the longest word in this sentence"]
|
|
)
|
|
assert scores[0]["interpretation"] == [
|
|
("Return", 0.0),
|
|
(" ", 0),
|
|
("the", 0.0),
|
|
(" ", 0),
|
|
("length", 0.0),
|
|
(" ", 0),
|
|
("of", 0.0),
|
|
(" ", 0),
|
|
("the", 0.0),
|
|
(" ", 0),
|
|
("longest", 0.0),
|
|
(" ", 0),
|
|
("word", 0.0),
|
|
(" ", 0),
|
|
("in", 0.0),
|
|
(" ", 0),
|
|
("this", 0.0),
|
|
(" ", 0),
|
|
("sentence", 1.0),
|
|
(" ", 0),
|
|
]
|
|
|
|
def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process
|
|
|
|
"""
|
|
iface = gr.Interface(lambda x: x[-1], "textbox", gr.Textbox())
|
|
assert iface("Hello") == "o"
|
|
iface = gr.Interface(lambda x: x / 2, "number", gr.Textbox())
|
|
assert iface(10) == "5.0"
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.Textbox("abc")
|
|
assert component.get_config().get("value") == "abc"
|
|
|
|
def test_override_template(self):
|
|
"""
|
|
override template
|
|
"""
|
|
component = gr.TextArea(value="abc")
|
|
assert component.get_config().get("value") == "abc"
|
|
assert component.get_config().get("lines") == 7
|
|
component = gr.TextArea(value="abc", lines=4)
|
|
assert component.get_config().get("value") == "abc"
|
|
assert component.get_config().get("lines") == 4
|
|
|
|
def test_faulty_type(self):
|
|
with pytest.raises(
|
|
ValueError, match='`type` must be one of "text", "password", or "email".'
|
|
):
|
|
gr.Textbox(type="boo")
|
|
|
|
def test_max_lines(self):
|
|
assert gr.Textbox(type="password").get_config().get("max_lines") == 1
|
|
assert gr.Textbox(type="email").get_config().get("max_lines") == 1
|
|
assert gr.Textbox(type="text").get_config().get("max_lines") == 20
|
|
assert gr.Textbox().get_config().get("max_lines") == 20
|
|
|
|
|
|
class TestNumber:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, set_interpret_parameters, get_interpretation_neighbors, get_config
|
|
|
|
"""
|
|
numeric_input = gr.Number()
|
|
assert numeric_input.preprocess(3) == 3.0
|
|
assert numeric_input.preprocess(None) is None
|
|
assert numeric_input.postprocess(3) == 3
|
|
assert numeric_input.postprocess(3) == 3.0
|
|
assert numeric_input.postprocess(2.14) == 2.14
|
|
assert numeric_input.postprocess(None) is None
|
|
assert numeric_input.serialize(3, True) == 3
|
|
assert isinstance(numeric_input.generate_sample(), float)
|
|
numeric_input.set_interpret_parameters(steps=3, delta=1, delta_type="absolute")
|
|
assert numeric_input.get_interpretation_neighbors(1) == (
|
|
[-2.0, -1.0, 0.0, 2.0, 3.0, 4.0],
|
|
{},
|
|
)
|
|
numeric_input.set_interpret_parameters(steps=3, delta=1, delta_type="percent")
|
|
assert numeric_input.get_interpretation_neighbors(1) == (
|
|
[0.97, 0.98, 0.99, 1.01, 1.02, 1.03],
|
|
{},
|
|
)
|
|
assert numeric_input.get_config() == {
|
|
"value": None,
|
|
"name": "number",
|
|
"show_label": True,
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
def test_component_functions_integer(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, set_interpret_parameters, get_interpretation_neighbors, get_template_context
|
|
|
|
"""
|
|
numeric_input = gr.Number(precision=0, value=42)
|
|
assert numeric_input.preprocess(3) == 3
|
|
assert numeric_input.preprocess(None) is None
|
|
assert numeric_input.postprocess(3) == 3
|
|
assert numeric_input.postprocess(3) == 3
|
|
assert numeric_input.postprocess(2.85) == 3
|
|
assert numeric_input.postprocess(None) is None
|
|
assert numeric_input.serialize(3, True) == 3
|
|
assert isinstance(numeric_input.generate_sample(), int)
|
|
numeric_input.set_interpret_parameters(steps=3, delta=1, delta_type="absolute")
|
|
assert numeric_input.get_interpretation_neighbors(1) == (
|
|
[-2.0, -1.0, 0.0, 2.0, 3.0, 4.0],
|
|
{},
|
|
)
|
|
numeric_input.set_interpret_parameters(steps=3, delta=1, delta_type="percent")
|
|
assert numeric_input.get_interpretation_neighbors(100) == (
|
|
[97.0, 98.0, 99.0, 101.0, 102.0, 103.0],
|
|
{},
|
|
)
|
|
with pytest.raises(ValueError) as error:
|
|
numeric_input.get_interpretation_neighbors(1)
|
|
assert error.msg == "Cannot generate valid set of neighbors"
|
|
numeric_input.set_interpret_parameters(
|
|
steps=3, delta=1.24, delta_type="absolute"
|
|
)
|
|
with pytest.raises(ValueError) as error:
|
|
numeric_input.get_interpretation_neighbors(4)
|
|
assert error.msg == "Cannot generate valid set of neighbors"
|
|
assert numeric_input.get_config() == {
|
|
"value": 42,
|
|
"name": "number",
|
|
"show_label": True,
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
def test_component_functions_precision(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, set_interpret_parameters, get_interpretation_neighbors, get_template_context
|
|
|
|
"""
|
|
numeric_input = gr.Number(precision=2, value=42.3428)
|
|
assert numeric_input.preprocess(3.231241) == 3.23
|
|
assert numeric_input.preprocess(None) is None
|
|
assert numeric_input.postprocess(-42.1241) == -42.12
|
|
assert numeric_input.postprocess(5.6784) == 5.68
|
|
assert numeric_input.postprocess(2.1421) == 2.14
|
|
assert numeric_input.postprocess(None) is None
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface_as_input(self):
|
|
"""
|
|
Interface, process, interpret
|
|
"""
|
|
iface = gr.Interface(lambda x: x**2, "number", "textbox")
|
|
assert iface(2) == "4.0"
|
|
iface = gr.Interface(
|
|
lambda x: x**2, "number", "number", interpretation="default"
|
|
)
|
|
scores = (await iface.interpret([2]))[0]["interpretation"]
|
|
assert scores == [
|
|
(1.94, -0.23640000000000017),
|
|
(1.96, -0.15840000000000032),
|
|
(1.98, -0.07960000000000012),
|
|
(2, None),
|
|
(2.02, 0.08040000000000003),
|
|
(2.04, 0.16159999999999997),
|
|
(2.06, 0.24359999999999982),
|
|
]
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_precision_0_in_interface(self):
|
|
"""
|
|
Interface, process, interpret
|
|
"""
|
|
iface = gr.Interface(lambda x: x**2, gr.Number(precision=0), "textbox")
|
|
assert iface(2) == "4"
|
|
iface = gr.Interface(
|
|
lambda x: x**2, "number", gr.Number(precision=0), interpretation="default"
|
|
)
|
|
# Output gets rounded to 4 for all input so no change
|
|
scores = (await iface.interpret([2]))[0]["interpretation"]
|
|
assert scores == [
|
|
(1.94, 0.0),
|
|
(1.96, 0.0),
|
|
(1.98, 0.0),
|
|
(2, None),
|
|
(2.02, 0.0),
|
|
(2.04, 0.0),
|
|
(2.06, 0.0),
|
|
]
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process, interpret
|
|
"""
|
|
iface = gr.Interface(lambda x: int(x) ** 2, "textbox", "number")
|
|
assert iface(2) == 4.0
|
|
iface = gr.Interface(
|
|
lambda x: x**2, "number", "number", interpretation="default"
|
|
)
|
|
scores = (await iface.interpret([2]))[0]["interpretation"]
|
|
assert scores == [
|
|
(1.94, -0.23640000000000017),
|
|
(1.96, -0.15840000000000032),
|
|
(1.98, -0.07960000000000012),
|
|
(2, None),
|
|
(2.02, 0.08040000000000003),
|
|
(2.04, 0.16159999999999997),
|
|
(2.06, 0.24359999999999982),
|
|
]
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.Number()
|
|
assert component.get_config().get("value") is None
|
|
component = gr.Number(3)
|
|
assert component.get_config().get("value") == 3.0
|
|
|
|
|
|
class TestSlider:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config
|
|
"""
|
|
slider_input = gr.Slider()
|
|
assert slider_input.preprocess(3.0) == 3.0
|
|
assert slider_input.postprocess(3) == 3
|
|
assert slider_input.postprocess(3) == 3
|
|
assert slider_input.postprocess(None) == 0
|
|
assert slider_input.serialize(3, True) == 3
|
|
|
|
assert isinstance(slider_input.generate_sample(), int)
|
|
slider_input = gr.Slider(10, 20, value=15, step=1, label="Slide Your Input")
|
|
assert slider_input.get_config() == {
|
|
"minimum": 10,
|
|
"maximum": 20,
|
|
"step": 1,
|
|
"value": 15,
|
|
"name": "slider",
|
|
"show_label": True,
|
|
"label": "Slide Your Input",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface(self):
|
|
""" "
|
|
Interface, process, interpret
|
|
"""
|
|
iface = gr.Interface(lambda x: x**2, "slider", "textbox")
|
|
assert iface(2) == "4"
|
|
iface = gr.Interface(
|
|
lambda x: x**2, "slider", "number", interpretation="default"
|
|
)
|
|
scores = (await iface.interpret([2]))[0]["interpretation"]
|
|
assert scores == [
|
|
-4.0,
|
|
200.08163265306123,
|
|
812.3265306122449,
|
|
1832.7346938775513,
|
|
3261.3061224489797,
|
|
5098.040816326531,
|
|
7342.938775510205,
|
|
9996.0,
|
|
]
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.Slider(0, 100, 5)
|
|
assert component.get_config().get("value") == 5
|
|
component = gr.Slider(0, 100, None)
|
|
assert component.get_config().get("value") == 0
|
|
|
|
@patch("gradio.Slider.get_random_value", return_value=7)
|
|
def test_slider_get_random_value_on_load(self, mock_get_random_value):
|
|
slider = gr.Slider(minimum=-5, maximum=10, randomize=True)
|
|
assert slider.value == 7
|
|
assert slider.load_event_to_attach[0]() == 7
|
|
assert slider.load_event_to_attach[1] is None
|
|
|
|
@patch("random.randint", return_value=3)
|
|
def test_slider_rounds_when_using_default_randomizer(self, mock_randint):
|
|
slider = gr.Slider(minimum=0, maximum=1, randomize=True, step=0.1)
|
|
# If get_random_value didn't round, this test would fail
|
|
# because 0.30000000000000004 != 0.3
|
|
assert slider.get_random_value() == 0.3
|
|
mock_randint.assert_called()
|
|
|
|
|
|
class TestCheckbox:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config
|
|
"""
|
|
bool_input = gr.Checkbox()
|
|
assert bool_input.preprocess(True)
|
|
assert bool_input.postprocess(True)
|
|
assert bool_input.postprocess(True)
|
|
assert bool_input.serialize(True, True)
|
|
assert isinstance(bool_input.generate_sample(), bool)
|
|
bool_input = gr.Checkbox(value=True, label="Check Your Input")
|
|
assert bool_input.get_config() == {
|
|
"value": True,
|
|
"name": "checkbox",
|
|
"show_label": True,
|
|
"label": "Check Your Input",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface(self):
|
|
"""
|
|
Interface, process, interpret
|
|
"""
|
|
iface = gr.Interface(lambda x: 1 if x else 0, "checkbox", "number")
|
|
assert iface(True) == 1
|
|
iface = gr.Interface(
|
|
lambda x: 1 if x else 0, "checkbox", "number", interpretation="default"
|
|
)
|
|
scores = (await iface.interpret([False]))[0]["interpretation"]
|
|
assert scores == (None, 1.0)
|
|
scores = (await iface.interpret([True]))[0]["interpretation"]
|
|
assert scores == (-1.0, None)
|
|
|
|
|
|
class TestCheckboxGroup:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config
|
|
"""
|
|
checkboxes_input = gr.CheckboxGroup(["a", "b", "c"])
|
|
assert checkboxes_input.preprocess(["a", "c"]) == ["a", "c"]
|
|
assert checkboxes_input.postprocess(["a", "c"]) == ["a", "c"]
|
|
assert checkboxes_input.serialize(["a", "c"], True) == ["a", "c"]
|
|
assert isinstance(checkboxes_input.generate_sample(), list)
|
|
checkboxes_input = gr.CheckboxGroup(
|
|
value=["a", "c"],
|
|
choices=["a", "b", "c"],
|
|
label="Check Your Inputs",
|
|
)
|
|
assert checkboxes_input.get_config() == {
|
|
"choices": ["a", "b", "c"],
|
|
"value": ["a", "c"],
|
|
"name": "checkboxgroup",
|
|
"show_label": True,
|
|
"label": "Check Your Inputs",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
with pytest.raises(ValueError):
|
|
gr.CheckboxGroup(["a"], type="unknown")
|
|
|
|
cbox = gr.CheckboxGroup(choices=["a", "b"], value="c")
|
|
assert cbox.get_config()["value"] == ["c"]
|
|
assert cbox.postprocess("a") == ["a"]
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
checkboxes_input = gr.CheckboxGroup(["a", "b", "c"])
|
|
iface = gr.Interface(lambda x: "|".join(x), checkboxes_input, "textbox")
|
|
assert iface(["a", "c"]) == "a|c"
|
|
assert iface([]) == ""
|
|
_ = gr.CheckboxGroup(["a", "b", "c"], type="index")
|
|
|
|
|
|
class TestRadio:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config
|
|
|
|
"""
|
|
radio_input = gr.Radio(["a", "b", "c"])
|
|
assert radio_input.preprocess("c") == "c"
|
|
assert radio_input.postprocess("a") == "a"
|
|
assert radio_input.serialize("a", True) == "a"
|
|
assert isinstance(radio_input.generate_sample(), str)
|
|
radio_input = gr.Radio(
|
|
choices=["a", "b", "c"], default="a", label="Pick Your One Input"
|
|
)
|
|
assert radio_input.get_config() == {
|
|
"choices": ["a", "b", "c"],
|
|
"value": None,
|
|
"name": "radio",
|
|
"show_label": True,
|
|
"label": "Pick Your One Input",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
with pytest.raises(ValueError):
|
|
gr.Radio(["a", "b"], type="unknown")
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface(self):
|
|
"""
|
|
Interface, process, interpret
|
|
"""
|
|
radio_input = gr.Radio(["a", "b", "c"])
|
|
iface = gr.Interface(lambda x: 2 * x, radio_input, "textbox")
|
|
assert iface("c") == "cc"
|
|
radio_input = gr.Radio(["a", "b", "c"], type="index")
|
|
iface = gr.Interface(
|
|
lambda x: 2 * x, radio_input, "number", interpretation="default"
|
|
)
|
|
assert iface("c") == 4
|
|
scores = (await iface.interpret(["b"]))[0]["interpretation"]
|
|
assert scores == [-2.0, None, 2.0]
|
|
|
|
|
|
class TestDropdown:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config
|
|
"""
|
|
dropdown_input = gr.Dropdown(["a", "b", "c"], multiselect=True)
|
|
assert dropdown_input.preprocess("a") == "a"
|
|
assert dropdown_input.postprocess("a") == "a"
|
|
|
|
dropdown_input_multiselect = gr.Dropdown(["a", "b", "c"], multiselect=True)
|
|
assert dropdown_input_multiselect.preprocess(["a", "c"]) == ["a", "c"]
|
|
assert dropdown_input_multiselect.postprocess(["a", "c"]) == ["a", "c"]
|
|
assert dropdown_input_multiselect.serialize(["a", "c"], True) == ["a", "c"]
|
|
assert isinstance(dropdown_input_multiselect.generate_sample(), str)
|
|
dropdown_input_multiselect = gr.Dropdown(
|
|
value=["a", "c"],
|
|
choices=["a", "b", "c"],
|
|
label="Select Your Inputs",
|
|
)
|
|
assert dropdown_input_multiselect.get_config() == {
|
|
"choices": ["a", "b", "c"],
|
|
"value": ["a", "c"],
|
|
"name": "dropdown",
|
|
"show_label": True,
|
|
"label": "Select Your Inputs",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"multiselect": None,
|
|
}
|
|
with pytest.raises(ValueError):
|
|
gr.Dropdown(["a"], type="unknown")
|
|
|
|
dropdown = gr.Dropdown(choices=["a", "b"], value="c")
|
|
assert dropdown.get_config()["value"] == "c"
|
|
assert dropdown.postprocess("a") == "a"
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
checkboxes_input = gr.CheckboxGroup(["a", "b", "c"])
|
|
iface = gr.Interface(lambda x: "|".join(x), checkboxes_input, "textbox")
|
|
assert iface(["a", "c"]) == "a|c"
|
|
assert iface([]) == ""
|
|
_ = gr.CheckboxGroup(["a", "b", "c"], type="index")
|
|
|
|
|
|
class TestImage:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, generate_sample, get_config, _segment_by_slic
|
|
type: pil, file, filepath, numpy
|
|
"""
|
|
img = deepcopy(media_data.BASE64_IMAGE)
|
|
image_input = gr.Image()
|
|
assert image_input.preprocess(img).shape == (68, 61, 3)
|
|
image_input = gr.Image(shape=(25, 25), image_mode="L")
|
|
assert image_input.preprocess(img).shape == (25, 25)
|
|
image_input = gr.Image(shape=(30, 10), type="pil")
|
|
assert image_input.preprocess(img).size == (30, 10)
|
|
assert image_input.postprocess("test/test_files/bus.png") == img
|
|
assert image_input.serialize("test/test_files/bus.png") == img
|
|
|
|
assert isinstance(image_input.generate_sample(), str)
|
|
image_input = gr.Image(
|
|
source="upload", tool="editor", type="pil", label="Upload Your Image"
|
|
)
|
|
assert image_input.get_config() == {
|
|
"image_mode": "RGB",
|
|
"shape": None,
|
|
"source": "upload",
|
|
"tool": "editor",
|
|
"name": "image",
|
|
"streaming": False,
|
|
"show_label": True,
|
|
"label": "Upload Your Image",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"mirror_webcam": True,
|
|
}
|
|
assert image_input.preprocess(None) is None
|
|
image_input = gr.Image(invert_colors=True)
|
|
assert image_input.preprocess(img) is not None
|
|
image_input.preprocess(img)
|
|
file_image = gr.Image(type="filepath")
|
|
assert isinstance(file_image.preprocess(img), str)
|
|
with pytest.raises(ValueError):
|
|
gr.Image(type="unknown")
|
|
image_input.shape = (30, 10)
|
|
assert image_input._segment_by_slic(img) is not None
|
|
|
|
# Output functionalities
|
|
y_img = gr.processing_utils.decode_base64_to_image(
|
|
deepcopy(media_data.BASE64_IMAGE)
|
|
)
|
|
image_output = gr.Image()
|
|
assert image_output.postprocess(y_img).startswith(
|
|
"data:image/png;base64,iVBORw0KGgoAAA"
|
|
)
|
|
assert image_output.postprocess(np.array(y_img)).startswith(
|
|
"data:image/png;base64,iVBORw0KGgoAAA"
|
|
)
|
|
with pytest.raises(ValueError):
|
|
image_output.postprocess([1, 2, 3])
|
|
image_output = gr.Image(type="numpy")
|
|
assert image_output.postprocess(y_img).startswith("data:image/png;base64,")
|
|
|
|
@pytest.mark.flaky
|
|
def test_serialize_url(self):
|
|
img = "https://gradio.app/assets/img/header-image.jpg"
|
|
expected = processing_utils.encode_url_or_file_to_base64(img)
|
|
assert gr.Image().serialize(img) == expected
|
|
|
|
def test_in_interface_as_input(self):
|
|
"""
|
|
Interface, process, interpret
|
|
type: file
|
|
interpretation: default, shap,
|
|
"""
|
|
img = "test/test_files/bus.png"
|
|
image_input = gr.Image()
|
|
iface = gr.Interface(
|
|
lambda x: PIL.Image.open(x).rotate(90, expand=True),
|
|
gr.Image(shape=(30, 10), type="filepath"),
|
|
"image",
|
|
)
|
|
output = iface(img)
|
|
assert PIL.Image.open(output).size == (10, 30)
|
|
iface = gr.Interface(
|
|
lambda x: np.sum(x), image_input, "number", interpretation="default"
|
|
)
|
|
|
|
def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def generate_noise(width, height):
|
|
return np.random.randint(0, 256, (width, height, 3))
|
|
|
|
iface = gr.Interface(generate_noise, ["slider", "slider"], "image")
|
|
assert iface(10, 20).endswith(".png")
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.Image("test/test_files/bus.png")
|
|
assert component.get_config().get("value") == media_data.BASE64_IMAGE
|
|
component = gr.Image(None)
|
|
assert component.get_config().get("value") is None
|
|
|
|
|
|
class TestPlot:
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def plot(num):
|
|
fig = plt.figure()
|
|
plt.plot(range(num), range(num))
|
|
return fig
|
|
|
|
iface = gr.Interface(plot, "slider", "plot")
|
|
output = await iface.process_api(fn_index=0, inputs=[10], state={})
|
|
assert output["data"][0]["type"] == "matplotlib"
|
|
assert output["data"][0]["plot"].startswith("data:image/png;base64")
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
fig = plt.figure()
|
|
plt.plot([1, 2, 3], [1, 2, 3])
|
|
|
|
component = gr.Plot(fig)
|
|
assert component.get_config().get("value") is not None
|
|
component = gr.Plot(None)
|
|
assert component.get_config().get("value") is None
|
|
|
|
def test_postprocess_altair(self):
|
|
import altair as alt
|
|
from vega_datasets import data
|
|
|
|
cars = data.cars()
|
|
chart = (
|
|
alt.Chart(cars)
|
|
.mark_point()
|
|
.encode(
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
color="Origin",
|
|
)
|
|
)
|
|
out = gr.Plot().postprocess(chart)
|
|
assert isinstance(out["plot"], str)
|
|
assert out["plot"] == chart.to_json()
|
|
|
|
|
|
class TestAudio:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess serialize, generate_sample, get_config, deserialize
|
|
type: filepath, numpy, file
|
|
"""
|
|
x_wav = deepcopy(media_data.BASE64_AUDIO)
|
|
audio_input = gr.Audio()
|
|
output = audio_input.preprocess(x_wav)
|
|
assert output[0] == 8000
|
|
assert output[1].shape == (8046,)
|
|
assert filecmp.cmp(
|
|
"test/test_files/audio_sample.wav",
|
|
audio_input.serialize("test/test_files/audio_sample.wav")["name"],
|
|
)
|
|
|
|
assert isinstance(audio_input.generate_sample(), dict)
|
|
audio_input = gr.Audio(label="Upload Your Audio")
|
|
assert audio_input.get_config() == {
|
|
"source": "upload",
|
|
"name": "audio",
|
|
"streaming": False,
|
|
"show_label": True,
|
|
"label": "Upload Your Audio",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
assert audio_input.preprocess(None) is None
|
|
x_wav["is_example"] = True
|
|
x_wav["crop_min"], x_wav["crop_max"] = 1, 4
|
|
assert audio_input.preprocess(x_wav) is not None
|
|
|
|
audio_input = gr.Audio(type="filepath")
|
|
assert isinstance(audio_input.preprocess(x_wav), str)
|
|
with pytest.raises(ValueError):
|
|
gr.Audio(type="unknown")
|
|
|
|
# Output functionalities
|
|
y_audio = gr.processing_utils.decode_base64_to_file(
|
|
deepcopy(media_data.BASE64_AUDIO)["data"]
|
|
)
|
|
audio_output = gr.Audio(type="filepath")
|
|
assert filecmp.cmp(y_audio.name, audio_output.postprocess(y_audio.name)["name"])
|
|
assert audio_output.get_config() == {
|
|
"name": "audio",
|
|
"streaming": False,
|
|
"show_label": True,
|
|
"label": None,
|
|
"source": "upload",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
assert audio_output.deserialize(
|
|
{
|
|
"name": None,
|
|
"data": deepcopy(media_data.BASE64_AUDIO)["data"],
|
|
"is_file": False,
|
|
}
|
|
).endswith(".wav")
|
|
|
|
output1 = audio_output.postprocess(y_audio.name)
|
|
output2 = audio_output.postprocess(y_audio.name)
|
|
assert output1 == output2
|
|
|
|
def test_serialize(self):
|
|
audio_input = gr.Audio()
|
|
assert audio_input.serialize("test/test_files/audio_sample.wav") == {
|
|
"data": media_data.BASE64_AUDIO["data"],
|
|
"is_file": False,
|
|
"orig_name": "audio_sample.wav",
|
|
"name": "test/test_files/audio_sample.wav",
|
|
}
|
|
|
|
def test_tokenize(self):
|
|
"""
|
|
Tokenize, get_masked_inputs
|
|
"""
|
|
x_wav = deepcopy(media_data.BASE64_AUDIO)
|
|
audio_input = gr.Audio()
|
|
tokens, _, _ = audio_input.tokenize(x_wav)
|
|
assert len(tokens) == audio_input.interpretation_segments
|
|
x_new = audio_input.get_masked_inputs(tokens, [[1] * len(tokens)])[0]
|
|
similarity = SequenceMatcher(a=x_wav["data"], b=x_new).ratio()
|
|
assert similarity > 0.9
|
|
|
|
def test_in_interface(self):
|
|
def reverse_audio(audio):
|
|
sr, data = audio
|
|
return (sr, np.flipud(data))
|
|
|
|
iface = gr.Interface(reverse_audio, "audio", "audio")
|
|
reversed_file = iface("test/test_files/audio_sample.wav")
|
|
reversed_reversed_file = iface(reversed_file)
|
|
reversed_reversed_data = gr.processing_utils.encode_url_or_file_to_base64(
|
|
reversed_reversed_file
|
|
)
|
|
similarity = SequenceMatcher(
|
|
a=reversed_reversed_data, b=media_data.BASE64_AUDIO["data"]
|
|
).ratio()
|
|
assert similarity > 0.99
|
|
|
|
def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def generate_noise(duration):
|
|
return 48000, np.random.randint(-256, 256, (duration, 3)).astype(np.int16)
|
|
|
|
iface = gr.Interface(generate_noise, "slider", "audio")
|
|
assert iface(100).endswith(".wav")
|
|
|
|
def test_audio_preprocess_can_be_read_by_scipy(self):
|
|
x_wav = deepcopy(media_data.BASE64_MICROPHONE)
|
|
audio_input = gr.Audio(type="filepath")
|
|
output = audio_input.preprocess(x_wav)
|
|
wavfile.read(output)
|
|
|
|
|
|
class TestFile:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, serialize, generate_sample, get_config, value
|
|
"""
|
|
x_file = deepcopy(media_data.BASE64_FILE)
|
|
file_input = gr.File()
|
|
output = file_input.preprocess(x_file)
|
|
assert isinstance(output, tempfile._TemporaryFileWrapper)
|
|
serialized = file_input.serialize("test/test_files/sample_file.pdf")
|
|
assert filecmp.cmp(
|
|
serialized["name"],
|
|
"test/test_files/sample_file.pdf",
|
|
)
|
|
assert serialized["orig_name"] == "sample_file.pdf"
|
|
assert output.orig_name == "test/test_files/sample_file.pdf"
|
|
|
|
x_file["is_file"] = True
|
|
input1 = file_input.preprocess(x_file)
|
|
input2 = file_input.preprocess(x_file)
|
|
assert input1.name == input2.name
|
|
|
|
assert isinstance(file_input.generate_sample(), dict)
|
|
file_input = gr.File(label="Upload Your File")
|
|
assert file_input.get_config() == {
|
|
"file_count": "single",
|
|
"file_types": None,
|
|
"name": "file",
|
|
"show_label": True,
|
|
"label": "Upload Your File",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
assert file_input.preprocess(None) is None
|
|
x_file["is_example"] = True
|
|
assert file_input.preprocess(x_file) is not None
|
|
|
|
zero_size_file = {"name": "document.txt", "size": 0, "data": "data:"}
|
|
temp_file = file_input.preprocess(zero_size_file)
|
|
assert os.stat(temp_file.name).st_size == 0
|
|
|
|
file_input = gr.File(type="binary")
|
|
output = file_input.preprocess(x_file)
|
|
assert type(output) == bytes
|
|
|
|
output1 = file_input.postprocess("test/test_files/sample_file.pdf")
|
|
output2 = file_input.postprocess("test/test_files/sample_file.pdf")
|
|
assert output1 == output2
|
|
|
|
def test_file_type_must_be_list(self):
|
|
with pytest.raises(
|
|
ValueError, match="Parameter file_types must be a list. Received str"
|
|
):
|
|
gr.File(file_types=".json")
|
|
|
|
def test_in_interface_as_input(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
x_file = media_data.BASE64_FILE["name"]
|
|
|
|
def get_size_of_file(file_obj):
|
|
return os.path.getsize(file_obj.name)
|
|
|
|
iface = gr.Interface(get_size_of_file, "file", "number")
|
|
assert iface(x_file) == 10558
|
|
|
|
def test_as_component_as_output(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def write_file(content):
|
|
with open("test.txt", "w") as f:
|
|
f.write(content)
|
|
return "test.txt"
|
|
|
|
iface = gr.Interface(write_file, "text", "file")
|
|
assert iface("hello world").endswith(".txt")
|
|
|
|
|
|
class TestUploadButton:
|
|
def test_component_functions(self):
|
|
"""
|
|
preprocess
|
|
"""
|
|
x_file = deepcopy(media_data.BASE64_FILE)
|
|
upload_input = gr.UploadButton()
|
|
input = upload_input.preprocess(x_file)
|
|
assert isinstance(input, tempfile._TemporaryFileWrapper)
|
|
|
|
x_file["is_file"] = True
|
|
input1 = upload_input.preprocess(x_file)
|
|
input2 = upload_input.preprocess(x_file)
|
|
assert input1.name == input2.name
|
|
|
|
def test_raises_if_file_types_is_not_list(self):
|
|
with pytest.raises(
|
|
ValueError, match="Parameter file_types must be a list. Received int"
|
|
):
|
|
gr.UploadButton(file_types=2)
|
|
|
|
|
|
class TestDataframe:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, serialize, generate_sample, get_config
|
|
"""
|
|
x_data = {
|
|
"data": [["Tim", 12, False], ["Jan", 24, True]],
|
|
"headers": ["Name", "Age", "Member"],
|
|
}
|
|
dataframe_input = gr.Dataframe(headers=["Name", "Age", "Member"])
|
|
output = dataframe_input.preprocess(x_data)
|
|
assert output["Age"][1] == 24
|
|
assert not output["Member"][0]
|
|
assert dataframe_input.postprocess(x_data) == x_data
|
|
|
|
assert isinstance(dataframe_input.generate_sample(), list)
|
|
dataframe_input = gr.Dataframe(
|
|
headers=["Name", "Age", "Member"], label="Dataframe Input"
|
|
)
|
|
assert dataframe_input.get_config() == {
|
|
"headers": ["Name", "Age", "Member"],
|
|
"datatype": ["str", "str", "str"],
|
|
"row_count": (1, "dynamic"),
|
|
"col_count": (3, "dynamic"),
|
|
"value": {
|
|
"data": [
|
|
["", "", ""],
|
|
],
|
|
"headers": ["Name", "Age", "Member"],
|
|
},
|
|
"name": "dataframe",
|
|
"show_label": True,
|
|
"label": "Dataframe Input",
|
|
"max_rows": 20,
|
|
"max_cols": None,
|
|
"overflow_row_behaviour": "paginate",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"wrap": False,
|
|
}
|
|
dataframe_input = gr.Dataframe()
|
|
output = dataframe_input.preprocess(x_data)
|
|
assert output["Age"][1] == 24
|
|
with pytest.raises(ValueError):
|
|
gr.Dataframe(type="unknown")
|
|
|
|
dataframe_output = gr.Dataframe()
|
|
assert dataframe_output.get_config() == {
|
|
"headers": [1, 2, 3],
|
|
"max_rows": 20,
|
|
"max_cols": None,
|
|
"overflow_row_behaviour": "paginate",
|
|
"name": "dataframe",
|
|
"show_label": True,
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"datatype": ["str", "str", "str"],
|
|
"row_count": (1, "dynamic"),
|
|
"col_count": (3, "dynamic"),
|
|
"value": {
|
|
"data": [
|
|
["", "", ""],
|
|
],
|
|
"headers": [1, 2, 3],
|
|
},
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"wrap": False,
|
|
}
|
|
|
|
def test_postprocess(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
dataframe_output = gr.Dataframe()
|
|
output = dataframe_output.postprocess(np.zeros((2, 2)))
|
|
assert output == {"data": [[0, 0], [0, 0]], "headers": [1, 2]}
|
|
output = dataframe_output.postprocess([[1, 3, 5]])
|
|
assert output == {"data": [[1, 3, 5]], "headers": [1, 2, 3]}
|
|
output = dataframe_output.postprocess(
|
|
pd.DataFrame([[2, True], [3, True], [4, False]], columns=["num", "prime"])
|
|
)
|
|
assert output == {
|
|
"headers": ["num", "prime"],
|
|
"data": [[2, True], [3, True], [4, False]],
|
|
}
|
|
with pytest.raises(ValueError):
|
|
gr.Dataframe(type="unknown")
|
|
|
|
# When the headers don't match the data
|
|
dataframe_output = gr.Dataframe(headers=["one", "two", "three"])
|
|
output = dataframe_output.postprocess([[2, True], [3, True]])
|
|
assert output == {
|
|
"headers": ["one", "two"],
|
|
"data": [[2, True], [3, True]],
|
|
}
|
|
dataframe_output = gr.Dataframe(headers=["one", "two", "three"])
|
|
output = dataframe_output.postprocess([[2, True, "ab", 4], [3, True, "cd", 5]])
|
|
assert output == {
|
|
"headers": ["one", "two", "three", 4],
|
|
"data": [[2, True, "ab", 4], [3, True, "cd", 5]],
|
|
}
|
|
|
|
def test_dataframe_postprocess_all_types(self):
|
|
df = pd.DataFrame(
|
|
{
|
|
"date_1": pd.date_range("2021-01-01", periods=2),
|
|
"date_2": pd.date_range("2022-02-15", periods=2).strftime(
|
|
"%B %d, %Y, %r"
|
|
),
|
|
"number": np.array([0.2233, 0.57281]),
|
|
"number_2": np.array([84, 23]).astype(np.int64),
|
|
"bool": [True, False],
|
|
"markdown": ["# Hello", "# Goodbye"],
|
|
}
|
|
)
|
|
component = gr.Dataframe(
|
|
datatype=["date", "date", "number", "number", "bool", "markdown"]
|
|
)
|
|
output = component.postprocess(df)
|
|
assert output == {
|
|
"headers": list(df.columns),
|
|
"data": [
|
|
[
|
|
pd.Timestamp("2021-01-01 00:00:00"),
|
|
"February 15, 2022, 12:00:00 AM",
|
|
0.2233,
|
|
84,
|
|
True,
|
|
"<h1>Hello</h1>\n",
|
|
],
|
|
[
|
|
pd.Timestamp("2021-01-02 00:00:00"),
|
|
"February 16, 2022, 12:00:00 AM",
|
|
0.57281,
|
|
23,
|
|
False,
|
|
"<h1>Goodbye</h1>\n",
|
|
],
|
|
],
|
|
}
|
|
|
|
def test_dataframe_postprocess_only_dates(self):
|
|
df = pd.DataFrame(
|
|
{
|
|
"date_1": pd.date_range("2021-01-01", periods=2),
|
|
"date_2": pd.date_range("2022-02-15", periods=2),
|
|
}
|
|
)
|
|
component = gr.Dataframe(datatype=["date", "date"])
|
|
output = component.postprocess(df)
|
|
assert output == {
|
|
"headers": list(df.columns),
|
|
"data": [
|
|
[
|
|
pd.Timestamp("2021-01-01 00:00:00"),
|
|
pd.Timestamp("2022-02-15 00:00:00"),
|
|
],
|
|
[
|
|
pd.Timestamp("2021-01-02 00:00:00"),
|
|
pd.Timestamp("2022-02-16 00:00:00"),
|
|
],
|
|
],
|
|
}
|
|
|
|
|
|
class TestDataset:
|
|
def test_preprocessing(self):
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
bus = str(pathlib.Path(test_file_dir, "bus.png").resolve())
|
|
|
|
dataset = gr.Dataset(
|
|
components=["number", "textbox", "image", "html", "markdown"],
|
|
samples=[
|
|
[5, "hello", bus, "<b>Bold</b>", "**Bold**"],
|
|
[15, "hi", bus, "<i>Italics</i>", "*Italics*"],
|
|
],
|
|
)
|
|
|
|
assert dataset.preprocess(1) == [
|
|
15,
|
|
"hi",
|
|
bus,
|
|
"<i>Italics</i>",
|
|
"<p><em>Italics</em></p>\n",
|
|
]
|
|
|
|
dataset = gr.Dataset(
|
|
components=["number", "textbox", "image", "html", "markdown"],
|
|
samples=[
|
|
[5, "hello", bus, "<b>Bold</b>", "**Bold**"],
|
|
[15, "hi", bus, "<i>Italics</i>", "*Italics*"],
|
|
],
|
|
type="index",
|
|
)
|
|
|
|
assert dataset.preprocess(1) == 1
|
|
|
|
def test_postprocessing(self):
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
bus = pathlib.Path(test_file_dir, "bus.png")
|
|
|
|
dataset = gr.Dataset(
|
|
components=["number", "textbox", "image", "html", "markdown"], type="index"
|
|
)
|
|
|
|
output = dataset.postprocess(
|
|
samples=[
|
|
[5, "hello", bus, "<b>Bold</b>", "**Bold**"],
|
|
[15, "hi", bus, "<i>Italics</i>", "*Italics*"],
|
|
],
|
|
)
|
|
|
|
assert output == {
|
|
"samples": [
|
|
[5, "hello", bus, "<b>Bold</b>", "**Bold**"],
|
|
[15, "hi", bus, "<i>Italics</i>", "*Italics*"],
|
|
],
|
|
"__type__": "update",
|
|
}
|
|
|
|
|
|
class TestVideo:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, serialize, deserialize, generate_sample, get_config
|
|
"""
|
|
x_video = deepcopy(media_data.BASE64_VIDEO)
|
|
video_input = gr.Video()
|
|
output1 = video_input.preprocess(x_video)
|
|
assert isinstance(output1, str)
|
|
output2 = video_input.preprocess(x_video)
|
|
assert output1 == output2
|
|
|
|
assert isinstance(video_input.generate_sample(), dict)
|
|
video_input = gr.Video(label="Upload Your Video")
|
|
assert video_input.get_config() == {
|
|
"source": "upload",
|
|
"name": "video",
|
|
"show_label": True,
|
|
"label": "Upload Your Video",
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"mirror_webcam": True,
|
|
"include_audio": True,
|
|
}
|
|
assert video_input.preprocess(None) is None
|
|
x_video["is_example"] = True
|
|
assert video_input.preprocess(x_video) is not None
|
|
video_input = gr.Video(format="avi")
|
|
output_video = video_input.preprocess(x_video)
|
|
assert output_video[-3:] == "avi"
|
|
assert "flip" not in output_video
|
|
|
|
assert filecmp.cmp(
|
|
video_input.serialize(x_video["name"])["name"], x_video["name"]
|
|
)
|
|
|
|
# Output functionalities
|
|
y_vid_path = "test/test_files/video_sample.mp4"
|
|
video_output = gr.Video()
|
|
output1 = video_output.postprocess(y_vid_path)["name"]
|
|
assert output1.endswith("mp4")
|
|
output2 = video_output.postprocess(y_vid_path)["name"]
|
|
assert output1 == output2
|
|
|
|
assert video_output.deserialize(
|
|
{
|
|
"name": None,
|
|
"data": deepcopy(media_data.BASE64_VIDEO)["data"],
|
|
"is_file": False,
|
|
}
|
|
).endswith(".mp4")
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
x_video = media_data.BASE64_VIDEO["name"]
|
|
iface = gr.Interface(lambda x: x, "video", "playable_video")
|
|
assert iface(x_video).endswith(".mp4")
|
|
|
|
def test_with_waveform(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
x_audio = media_data.BASE64_AUDIO["name"]
|
|
iface = gr.Interface(lambda x: gr.make_waveform(x), "audio", "video")
|
|
assert iface(x_audio).endswith(".mp4")
|
|
|
|
def test_video_postprocess_converts_to_playable_format(self):
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
# This file has a playable container but not playable codec
|
|
with tempfile.NamedTemporaryFile(
|
|
suffix="bad_video.mp4", delete=False
|
|
) as tmp_not_playable_vid:
|
|
bad_vid = str(test_file_dir / "bad_video_sample.mp4")
|
|
assert not processing_utils.video_is_playable(bad_vid)
|
|
shutil.copy(bad_vid, tmp_not_playable_vid.name)
|
|
_ = gr.Video().postprocess(tmp_not_playable_vid.name)
|
|
# The original video gets converted to .mp4 format
|
|
full_path_to_output = pathlib.Path(tmp_not_playable_vid.name).with_suffix(
|
|
".mp4"
|
|
)
|
|
assert processing_utils.video_is_playable(str(full_path_to_output))
|
|
|
|
# This file has a playable codec but not a playable container
|
|
with tempfile.NamedTemporaryFile(
|
|
suffix="playable_but_bad_container.mkv", delete=False
|
|
) as tmp_not_playable_vid:
|
|
bad_vid = str(test_file_dir / "playable_but_bad_container.mkv")
|
|
assert not processing_utils.video_is_playable(bad_vid)
|
|
shutil.copy(bad_vid, tmp_not_playable_vid.name)
|
|
_ = gr.Video().postprocess(tmp_not_playable_vid.name)
|
|
full_path_to_output = pathlib.Path(tmp_not_playable_vid.name).with_suffix(
|
|
".mp4"
|
|
)
|
|
assert processing_utils.video_is_playable(str(full_path_to_output))
|
|
|
|
@patch("pathlib.Path.exists", MagicMock(return_value=False))
|
|
@patch("gradio.components.FFmpeg")
|
|
def test_video_preprocessing_flips_video_for_webcam(self, mock_ffmpeg):
|
|
# Ensures that the cached temp video file is not used so that ffmpeg is called for each test
|
|
x_video = deepcopy(media_data.BASE64_VIDEO)
|
|
video_input = gr.Video(source="webcam")
|
|
_ = video_input.preprocess(x_video)
|
|
|
|
# Dict mapping filename to FFmpeg options
|
|
output_params = mock_ffmpeg.call_args_list[0][1]["outputs"]
|
|
assert "hflip" in list(output_params.values())[0]
|
|
assert "flip" in list(output_params.keys())[0]
|
|
|
|
mock_ffmpeg.reset_mock()
|
|
_ = gr.Video(
|
|
source="webcam", mirror_webcam=False, include_audio=True
|
|
).preprocess(x_video)
|
|
mock_ffmpeg.assert_not_called()
|
|
|
|
mock_ffmpeg.reset_mock()
|
|
_ = gr.Video(source="upload", format="mp4", include_audio=True).preprocess(
|
|
x_video
|
|
)
|
|
mock_ffmpeg.assert_not_called()
|
|
|
|
mock_ffmpeg.reset_mock()
|
|
output_file = gr.Video(
|
|
source="webcam", mirror_webcam=True, format="avi"
|
|
).preprocess(x_video)
|
|
output_params = mock_ffmpeg.call_args_list[0][1]["outputs"]
|
|
assert "hflip" in list(output_params.values())[0]
|
|
assert "flip" in list(output_params.keys())[0]
|
|
assert ".avi" in list(output_params.keys())[0]
|
|
assert ".avi" in output_file
|
|
|
|
mock_ffmpeg.reset_mock()
|
|
output_file = gr.Video(
|
|
source="webcam", mirror_webcam=False, format="avi", include_audio=False
|
|
).preprocess(x_video)
|
|
output_params = mock_ffmpeg.call_args_list[0][1]["outputs"]
|
|
assert list(output_params.values())[0] == ["-an"]
|
|
assert "flip" not in list(output_params.keys())[0]
|
|
assert ".avi" in list(output_params.keys())[0]
|
|
assert ".avi" in output_file
|
|
|
|
|
|
class TestTimeseries:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, generate_sample, get_config,
|
|
"""
|
|
timeseries_input = gr.Timeseries(x="time", y=["retail", "food", "other"])
|
|
x_timeseries = {
|
|
"data": [[1] + [2] * len(timeseries_input.y)] * 4,
|
|
"headers": [timeseries_input.x] + timeseries_input.y,
|
|
}
|
|
output = timeseries_input.preprocess(x_timeseries)
|
|
assert isinstance(output, pd.core.frame.DataFrame)
|
|
|
|
assert isinstance(timeseries_input.generate_sample(), dict)
|
|
timeseries_input = gr.Timeseries(
|
|
x="time", y="retail", label="Upload Your Timeseries"
|
|
)
|
|
assert timeseries_input.get_config() == {
|
|
"x": "time",
|
|
"y": ["retail"],
|
|
"name": "timeseries",
|
|
"show_label": True,
|
|
"label": "Upload Your Timeseries",
|
|
"colors": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
assert timeseries_input.preprocess(None) is None
|
|
x_timeseries["range"] = (0, 1)
|
|
assert timeseries_input.preprocess(x_timeseries) is not None
|
|
|
|
# Output functionalities
|
|
|
|
timeseries_output = gr.Timeseries(label="Disease")
|
|
|
|
assert timeseries_output.get_config() == {
|
|
"x": None,
|
|
"y": None,
|
|
"name": "timeseries",
|
|
"show_label": True,
|
|
"label": "Disease",
|
|
"colors": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
data = {"Name": ["Tom", "nick", "krish", "jack"], "Age": [20, 21, 19, 18]}
|
|
df = pd.DataFrame(data)
|
|
assert timeseries_output.postprocess(df) == {
|
|
"headers": ["Name", "Age"],
|
|
"data": [["Tom", 20], ["nick", 21], ["krish", 19], ["jack", 18]],
|
|
}
|
|
|
|
timeseries_output = gr.Timeseries(y="Age", label="Disease")
|
|
output = timeseries_output.postprocess(df)
|
|
assert output == {
|
|
"headers": ["Name", "Age"],
|
|
"data": [["Tom", 20], ["nick", 21], ["krish", 19], ["jack", 18]],
|
|
}
|
|
|
|
|
|
class TestNames:
|
|
# This test ensures that `components.get_component_instance()` works correctly when instantiating from components
|
|
def test_no_duplicate_uncased_names(self):
|
|
subclasses = gr.components.Component.__subclasses__()
|
|
unique_subclasses_uncased = set([s.__name__.lower() for s in subclasses])
|
|
assert len(subclasses) == len(unique_subclasses_uncased)
|
|
|
|
|
|
class TestLabel:
|
|
def test_component_functions(self):
|
|
"""
|
|
Process, postprocess, deserialize
|
|
"""
|
|
y = "happy"
|
|
label_output = gr.Label()
|
|
label = label_output.postprocess(y)
|
|
assert label == {"label": "happy"}
|
|
assert json.load(open(label_output.deserialize(label))) == label
|
|
|
|
y = {3: 0.7, 1: 0.2, 0: 0.1}
|
|
label = label_output.postprocess(y)
|
|
assert label == {
|
|
"label": 3,
|
|
"confidences": [
|
|
{"label": 3, "confidence": 0.7},
|
|
{"label": 1, "confidence": 0.2},
|
|
{"label": 0, "confidence": 0.1},
|
|
],
|
|
}
|
|
label_output = gr.Label(num_top_classes=2)
|
|
label = label_output.postprocess(y)
|
|
|
|
assert label == {
|
|
"label": 3,
|
|
"confidences": [
|
|
{"label": 3, "confidence": 0.7},
|
|
{"label": 1, "confidence": 0.2},
|
|
],
|
|
}
|
|
with pytest.raises(ValueError):
|
|
label_output.postprocess([1, 2, 3])
|
|
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
path = str(pathlib.Path(test_file_dir, "test_label_json.json"))
|
|
label_dict = label_output.postprocess(path)
|
|
assert label_dict["label"] == "web site"
|
|
|
|
assert label_output.get_config() == {
|
|
"name": "label",
|
|
"show_label": True,
|
|
"num_top_classes": 2,
|
|
"value": None,
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"color": None,
|
|
}
|
|
|
|
def test_color_argument(self):
|
|
|
|
label = gr.Label(value=-10, color="red")
|
|
assert label.get_config()["color"] == "red"
|
|
update_1 = gr.Label.update(value="bad", color="brown")
|
|
assert update_1["color"] == "brown"
|
|
update_2 = gr.Label.update(value="bad", color="#ff9966")
|
|
assert update_2["color"] == "#ff9966"
|
|
|
|
update_3 = gr.Label.update(
|
|
value={"bad": 0.9, "good": 0.09, "so-so": 0.01}, color="green"
|
|
)
|
|
assert update_3["color"] == "green"
|
|
|
|
update_4 = gr.Label.update(value={"bad": 0.8, "good": 0.18, "so-so": 0.02})
|
|
assert update_4["color"] is None
|
|
|
|
update_5 = gr.Label.update(
|
|
value={"bad": 0.8, "good": 0.18, "so-so": 0.02}, color=None
|
|
)
|
|
assert update_5["color"] == "transparent"
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
x_img = "test/test_files/bus.png"
|
|
|
|
def rgb_distribution(img):
|
|
rgb_dist = np.mean(img, axis=(0, 1))
|
|
rgb_dist /= np.sum(rgb_dist)
|
|
rgb_dist = np.round(rgb_dist, decimals=2)
|
|
return {
|
|
"red": rgb_dist[0],
|
|
"green": rgb_dist[1],
|
|
"blue": rgb_dist[2],
|
|
}
|
|
|
|
iface = gr.Interface(rgb_distribution, "image", "label")
|
|
output_filepath = iface(x_img)
|
|
with open(output_filepath) as fp:
|
|
assert json.load(fp) == {
|
|
"label": "red",
|
|
"confidences": [
|
|
{"label": "red", "confidence": 0.44},
|
|
{"label": "green", "confidence": 0.28},
|
|
{"label": "blue", "confidence": 0.28},
|
|
],
|
|
}
|
|
|
|
|
|
class TestHighlightedText:
|
|
def test_postprocess(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.HighlightedText()
|
|
result = [
|
|
("", None),
|
|
("Wolfgang", "PER"),
|
|
(" lives in ", None),
|
|
("Berlin", "LOC"),
|
|
("", None),
|
|
]
|
|
result_ = component.postprocess(result)
|
|
assert result == result_
|
|
|
|
text = "Wolfgang lives in Berlin"
|
|
entities = [
|
|
{"entity": "PER", "start": 0, "end": 8},
|
|
{"entity": "LOC", "start": 18, "end": 24},
|
|
]
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert result == result_
|
|
|
|
# Test split entity is merged when combine adjacent is set
|
|
text = "Wolfgang lives in Berlin"
|
|
entities = [
|
|
{"entity": "PER", "start": 0, "end": 4},
|
|
{"entity": "PER", "start": 4, "end": 8},
|
|
{"entity": "LOC", "start": 18, "end": 24},
|
|
]
|
|
# After a merge empty entries are stripped except the leading one
|
|
result_after_merge = [
|
|
("", None),
|
|
("Wolfgang", "PER"),
|
|
(" lives in ", None),
|
|
("Berlin", "LOC"),
|
|
]
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert result != result_
|
|
assert result_after_merge != result_
|
|
|
|
component = gr.HighlightedText(combine_adjacent=True)
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert result_after_merge == result_
|
|
|
|
component = gr.HighlightedText()
|
|
|
|
text = "Wolfgang lives in Berlin"
|
|
entities = [
|
|
{"entity": "LOC", "start": 18, "end": 24},
|
|
{"entity": "PER", "start": 0, "end": 8},
|
|
]
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert result == result_
|
|
|
|
text = "I live there"
|
|
entities = []
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert [(text, None)] == result_
|
|
|
|
text = "Wolfgang"
|
|
entities = [
|
|
{"entity": "PER", "start": 0, "end": 8},
|
|
]
|
|
result_ = component.postprocess({"text": text, "entities": entities})
|
|
assert [("", None), (text, "PER"), ("", None)] == result_
|
|
|
|
def test_component_functions(self):
|
|
"""
|
|
get_config
|
|
"""
|
|
ht_output = gr.HighlightedText(color_map={"pos": "green", "neg": "red"})
|
|
assert ht_output.get_config() == {
|
|
"color_map": {"pos": "green", "neg": "red"},
|
|
"name": "highlightedtext",
|
|
"show_label": True,
|
|
"label": None,
|
|
"show_legend": False,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def highlight_vowels(sentence):
|
|
phrases, cur_phrase = [], ""
|
|
vowels, mode = "aeiou", None
|
|
for letter in sentence:
|
|
letter_mode = "vowel" if letter in vowels else "non"
|
|
if mode is None:
|
|
mode = letter_mode
|
|
elif mode != letter_mode:
|
|
phrases.append((cur_phrase, mode))
|
|
cur_phrase = ""
|
|
mode = letter_mode
|
|
cur_phrase += letter
|
|
phrases.append((cur_phrase, mode))
|
|
return phrases
|
|
|
|
iface = gr.Interface(highlight_vowels, "text", "highlight")
|
|
output_filepath = iface("Helloooo")
|
|
with open(output_filepath) as fp:
|
|
output = json.load(fp)
|
|
assert output == [
|
|
["H", "non"],
|
|
["e", "vowel"],
|
|
["ll", "non"],
|
|
["oooo", "vowel"],
|
|
]
|
|
|
|
|
|
class TestChatbot:
|
|
def test_component_functions(self):
|
|
"""
|
|
Postprocess, get_config
|
|
"""
|
|
chatbot = gr.Chatbot()
|
|
assert chatbot.postprocess([("You are **cool**", "so are *you*")]) == [
|
|
("You are <strong>cool</strong>", "so are <em>you</em>")
|
|
]
|
|
assert chatbot.get_config() == {
|
|
"value": [],
|
|
"color_map": None,
|
|
"label": None,
|
|
"show_label": True,
|
|
"interactive": None,
|
|
"name": "chatbot",
|
|
"visible": True,
|
|
"elem_id": None,
|
|
"style": {},
|
|
"root_url": None,
|
|
}
|
|
|
|
|
|
class TestJSON:
|
|
def test_component_functions(self):
|
|
"""
|
|
Postprocess
|
|
"""
|
|
js_output = gr.JSON()
|
|
assert js_output.postprocess('{"a":1, "b": 2}'), '"{\\"a\\":1, \\"b\\": 2}"'
|
|
assert js_output.get_config() == {
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": None,
|
|
"show_label": True,
|
|
"label": None,
|
|
"name": "json",
|
|
"interactive": None,
|
|
"root_url": None,
|
|
}
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def get_avg_age_per_gender(data):
|
|
print(data)
|
|
return {
|
|
"M": int(data[data["gender"] == "M"].mean()),
|
|
"F": int(data[data["gender"] == "F"].mean()),
|
|
"O": int(data[data["gender"] == "O"].mean()),
|
|
}
|
|
|
|
iface = gr.Interface(
|
|
get_avg_age_per_gender,
|
|
gr.Dataframe(headers=["gender", "age"]),
|
|
"json",
|
|
)
|
|
y_data = [
|
|
["M", 30],
|
|
["F", 20],
|
|
["M", 40],
|
|
["O", 20],
|
|
["F", 30],
|
|
]
|
|
assert (
|
|
await iface.process_api(
|
|
0, [{"data": y_data, "headers": ["gender", "age"]}], state={}
|
|
)
|
|
)["data"][0] == {
|
|
"M": 35,
|
|
"F": 25,
|
|
"O": 20,
|
|
}
|
|
|
|
|
|
class TestHTML:
|
|
def test_component_functions(self):
|
|
"""
|
|
get_config
|
|
"""
|
|
html_component = gr.components.HTML("#Welcome onboard", label="HTML Input")
|
|
assert {
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"value": "#Welcome onboard",
|
|
"show_label": True,
|
|
"label": "HTML Input",
|
|
"name": "html",
|
|
"interactive": None,
|
|
"root_url": None,
|
|
} == html_component.get_config()
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
|
|
def bold_text(text):
|
|
return "<strong>" + text + "</strong>"
|
|
|
|
iface = gr.Interface(bold_text, "text", "html")
|
|
assert iface("test") == "<strong>test</strong>"
|
|
|
|
|
|
class TestMarkdown:
|
|
def test_component_functions(self):
|
|
markdown_component = gr.Markdown("# Let's learn about $x$", label="Markdown")
|
|
assert markdown_component.get_config()["value"].startswith(
|
|
"""<h1>Let\'s learn about <span class="math inline"><span style=\'font-size: 0px\'>x</span><svg xmlns:xlink="http://www.w3.org/1999/xlink" width="11.6pt" height="19.35625pt" viewBox="0 0 11.6 19.35625" xmlns="http://www.w3.org/2000/svg" version="1.1">\n \n <defs>\n <style type="text/css">*{stroke-linejoin: round; stroke-linecap: butt}</style>\n </defs>\n <g id="figure_1">\n <g id="patch_1">\n <path d="M 0 19.35625"""
|
|
)
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
iface = gr.Interface(lambda x: x, "text", "markdown")
|
|
input_data = "Here's an [image](https://gradio.app/images/gradio_logo.png)"
|
|
output_data = iface(input_data)
|
|
assert (
|
|
output_data
|
|
== """<p>Here's an <a href="https://gradio.app/images/gradio_logo.png">image</a></p>\n"""
|
|
)
|
|
|
|
|
|
class TestModel3D:
|
|
def test_component_functions(self):
|
|
"""
|
|
get_config
|
|
"""
|
|
component = gr.components.Model3D(None, label="Model")
|
|
assert {
|
|
"clearColor": [0, 0, 0, 0],
|
|
"value": None,
|
|
"label": "Model",
|
|
"show_label": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"name": "model3d",
|
|
"visible": True,
|
|
"elem_id": None,
|
|
"style": {},
|
|
} == component.get_config()
|
|
|
|
file = "test/test_files/Box.gltf"
|
|
output1 = component.postprocess(file)
|
|
output2 = component.postprocess(file)
|
|
assert output1 == output2
|
|
|
|
def test_in_interface(self):
|
|
"""
|
|
Interface, process
|
|
"""
|
|
iface = gr.Interface(lambda x: x, "model3d", "model3d")
|
|
input_data = "test/test_files/Box.gltf"
|
|
output_data = iface(input_data)
|
|
assert output_data.endswith(".gltf")
|
|
|
|
|
|
class TestColorPicker:
|
|
def test_component_functions(self):
|
|
"""
|
|
Preprocess, postprocess, serialize, tokenize, generate_sample, get_config
|
|
"""
|
|
color_picker_input = gr.ColorPicker()
|
|
assert color_picker_input.preprocess("#000000") == "#000000"
|
|
assert color_picker_input.postprocess("#000000") == "#000000"
|
|
assert color_picker_input.postprocess(None) is None
|
|
assert color_picker_input.postprocess("#FFFFFF") == "#FFFFFF"
|
|
assert color_picker_input.serialize("#000000", True) == "#000000"
|
|
|
|
color_picker_input.interpretation_replacement = "unknown"
|
|
|
|
assert color_picker_input.get_config() == {
|
|
"value": None,
|
|
"show_label": True,
|
|
"label": None,
|
|
"style": {},
|
|
"elem_id": None,
|
|
"visible": True,
|
|
"interactive": None,
|
|
"root_url": None,
|
|
"name": "colorpicker",
|
|
}
|
|
assert isinstance(color_picker_input.generate_sample(), str)
|
|
|
|
def test_in_interface_as_input(self):
|
|
"""
|
|
Interface, process, interpret,
|
|
"""
|
|
iface = gr.Interface(lambda x: x, "colorpicker", "colorpicker")
|
|
assert iface("#000000") == "#000000"
|
|
|
|
def test_in_interface_as_output(self):
|
|
"""
|
|
Interface, process
|
|
|
|
"""
|
|
iface = gr.Interface(lambda x: x, "colorpicker", gr.ColorPicker())
|
|
assert iface("#000000") == "#000000"
|
|
|
|
def test_static(self):
|
|
"""
|
|
postprocess
|
|
"""
|
|
component = gr.ColorPicker("#000000")
|
|
assert component.get_config().get("value") == "#000000"
|
|
|
|
|
|
class TestCarousel:
|
|
def test_deprecation(self):
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
with pytest.raises(DeprecationWarning):
|
|
gr.Carousel([pathlib.Path(test_file_dir, "bus.png")])
|
|
|
|
def test_deprecation_in_interface(self):
|
|
with pytest.raises(DeprecationWarning):
|
|
gr.Interface(lambda x: ["lion.jpg"], "textbox", "carousel")
|
|
|
|
def test_deprecation_in_blocks(self):
|
|
with pytest.raises(DeprecationWarning):
|
|
with gr.Blocks():
|
|
gr.Textbox()
|
|
gr.Carousel()
|
|
|
|
|
|
class TestGallery:
|
|
@patch("uuid.uuid4", return_value="my-uuid")
|
|
def test_gallery(self, mock_uuid):
|
|
gallery = gr.Gallery()
|
|
test_file_dir = pathlib.Path(pathlib.Path(__file__).parent, "test_files")
|
|
data = [
|
|
gr.processing_utils.encode_file_to_base64(
|
|
pathlib.Path(test_file_dir, "bus.png")
|
|
),
|
|
gr.processing_utils.encode_file_to_base64(
|
|
pathlib.Path(test_file_dir, "cheetah1.jpg")
|
|
),
|
|
]
|
|
|
|
with tempfile.TemporaryDirectory() as tmpdir:
|
|
path = gallery.deserialize(data, tmpdir)
|
|
assert path.endswith("my-uuid")
|
|
data_restored = gallery.serialize(path)
|
|
data_restored = [d[0]["data"] for d in data_restored]
|
|
assert sorted(data) == sorted(data_restored)
|
|
|
|
|
|
class TestState:
|
|
def test_as_component(self):
|
|
state = gr.State(value=5)
|
|
assert state.preprocess(10) == 10
|
|
assert state.preprocess("abc") == "abc"
|
|
assert state.stateful
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_interface(self):
|
|
def test(x, y=" def"):
|
|
return (x + y, x + y)
|
|
|
|
io = gr.Interface(test, ["text", "state"], ["text", "state"])
|
|
result = await io.call_function(0, ["abc"])
|
|
assert result["prediction"][0] == "abc def"
|
|
result = await io.call_function(0, ["abc", result["prediction"][0]])
|
|
assert result["prediction"][0] == "abcabc def"
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_in_blocks(self):
|
|
with gr.Blocks() as demo:
|
|
score = gr.State()
|
|
btn = gr.Button()
|
|
btn.click(lambda x: x + 1, score, score)
|
|
|
|
result = await demo.call_function(0, [0])
|
|
assert result["prediction"] == 1
|
|
result = await demo.call_function(0, [result["prediction"]])
|
|
assert result["prediction"] == 2
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_variable_for_backwards_compatibility(self):
|
|
with gr.Blocks() as demo:
|
|
score = gr.Variable()
|
|
btn = gr.Button()
|
|
btn.click(lambda x: x + 1, score, score)
|
|
|
|
result = await demo.call_function(0, [0])
|
|
assert result["prediction"] == 1
|
|
result = await demo.call_function(0, [result["prediction"]])
|
|
assert result["prediction"] == 2
|
|
|
|
|
|
def test_dataframe_as_example_converts_dataframes():
|
|
df_comp = gr.Dataframe()
|
|
assert df_comp.as_example(pd.DataFrame({"a": [1, 2, 3, 4], "b": [5, 6, 7, 8]})) == [
|
|
[1, 5],
|
|
[2, 6],
|
|
[3, 7],
|
|
[4, 8],
|
|
]
|
|
assert df_comp.as_example(np.array([[1, 2], [3, 4.0]])) == [[1.0, 2.0], [3.0, 4.0]]
|
|
|
|
|
|
@pytest.mark.parametrize("component", [gr.Model3D, gr.File, gr.Audio])
|
|
def test_as_example_returns_file_basename(component):
|
|
component = component()
|
|
assert component.as_example("/home/freddy/sources/example.ext") == "example.ext"
|
|
assert component.as_example(None) == ""
|
|
|
|
|
|
@patch("gradio.components.IOComponent.as_example")
|
|
@patch("gradio.components.Image.as_example")
|
|
@patch("gradio.components.File.as_example")
|
|
@patch("gradio.components.Dataframe.as_example")
|
|
@patch("gradio.components.Model3D.as_example")
|
|
def test_dataset_calls_as_example(*mocks):
|
|
gr.Dataset(
|
|
components=[gr.Dataframe(), gr.File(), gr.Image(), gr.Model3D(), gr.Textbox()],
|
|
samples=[
|
|
[
|
|
pd.DataFrame({"a": np.array([1, 2, 3])}),
|
|
"foo.png",
|
|
"bar.jpeg",
|
|
"duck.obj",
|
|
"hello",
|
|
]
|
|
],
|
|
)
|
|
assert all([m.called for m in mocks])
|
|
|
|
|
|
cars = vega_datasets.data.cars()
|
|
stocks = vega_datasets.data.stocks()
|
|
|
|
|
|
class TestScatterPlot:
|
|
def test_get_config(self):
|
|
assert gr.ScatterPlot().get_config() == {
|
|
"caption": None,
|
|
"elem_id": None,
|
|
"interactive": None,
|
|
"label": None,
|
|
"name": "plot",
|
|
"root_url": None,
|
|
"show_label": True,
|
|
"style": {},
|
|
"value": None,
|
|
"visible": True,
|
|
}
|
|
|
|
def test_no_color(self):
|
|
plot = gr.ScatterPlot(
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
tooltip="Name",
|
|
title="Car Data",
|
|
x_title="Horse",
|
|
)
|
|
output = plot.postprocess(cars)
|
|
assert sorted(list(output.keys())) == ["chart", "plot", "type"]
|
|
config = json.loads(output["plot"])
|
|
assert config["encoding"]["x"]["field"] == "Horsepower"
|
|
assert config["encoding"]["x"]["title"] == "Horse"
|
|
assert config["encoding"]["y"]["field"] == "Miles_per_Gallon"
|
|
assert config["selection"] == {
|
|
"selector001": {
|
|
"bind": "scales",
|
|
"encodings": ["x", "y"],
|
|
"type": "interval",
|
|
}
|
|
}
|
|
assert config["title"] == "Car Data"
|
|
assert "height" not in config
|
|
assert "width" not in config
|
|
|
|
def test_no_interactive(self):
|
|
plot = gr.ScatterPlot(
|
|
x="Horsepower", y="Miles_per_Gallon", tooltip="Name", interactive=False
|
|
)
|
|
output = plot.postprocess(cars)
|
|
assert sorted(list(output.keys())) == ["chart", "plot", "type"]
|
|
config = json.loads(output["plot"])
|
|
assert "selection" not in config
|
|
|
|
def test_height_width(self):
|
|
plot = gr.ScatterPlot(
|
|
x="Horsepower", y="Miles_per_Gallon", height=100, width=200
|
|
)
|
|
output = plot.postprocess(cars)
|
|
assert sorted(list(output.keys())) == ["chart", "plot", "type"]
|
|
config = json.loads(output["plot"])
|
|
assert config["height"] == 100
|
|
assert config["width"] == 200
|
|
|
|
def test_xlim_ylim(self):
|
|
plot = gr.ScatterPlot(
|
|
x="Horsepower", y="Miles_per_Gallon", x_lim=[200, 400], y_lim=[300, 500]
|
|
)
|
|
output = plot.postprocess(cars)
|
|
config = json.loads(output["plot"])
|
|
assert config["encoding"]["x"]["scale"] == {"domain": [200, 400]}
|
|
assert config["encoding"]["y"]["scale"] == {"domain": [300, 500]}
|
|
|
|
def test_color_encoding(self):
|
|
plot = gr.ScatterPlot(
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
tooltip="Name",
|
|
title="Car Data",
|
|
color="Origin",
|
|
)
|
|
output = plot.postprocess(cars)
|
|
config = json.loads(output["plot"])
|
|
assert config["encoding"]["color"]["field"] == "Origin"
|
|
assert config["encoding"]["color"]["scale"] == {
|
|
"domain": ["USA", "Europe", "Japan"],
|
|
"range": [0, 1, 2],
|
|
}
|
|
assert config["encoding"]["color"]["type"] == "nominal"
|
|
|
|
def test_two_encodings(self):
|
|
plot = gr.ScatterPlot(
|
|
show_label=False,
|
|
title="Two encodings",
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
color="Acceleration",
|
|
shape="Origin",
|
|
)
|
|
output = plot.postprocess(cars)
|
|
config = json.loads(output["plot"])
|
|
assert config["encoding"]["color"]["field"] == "Acceleration"
|
|
assert config["encoding"]["color"]["scale"] == {
|
|
"domain": [cars.Acceleration.min(), cars.Acceleration.max()],
|
|
"range": [0, 1],
|
|
}
|
|
assert config["encoding"]["color"]["type"] == "quantitative"
|
|
|
|
assert config["encoding"]["shape"]["field"] == "Origin"
|
|
assert config["encoding"]["shape"]["type"] == "nominal"
|
|
|
|
def test_legend_position(self):
|
|
plot = gr.ScatterPlot(
|
|
show_label=False,
|
|
title="Two encodings",
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
color="Acceleration",
|
|
color_legend_position="none",
|
|
color_legend_title="Foo",
|
|
shape="Origin",
|
|
shape_legend_position="none",
|
|
shape_legend_title="Bar",
|
|
size="Acceleration",
|
|
size_legend_title="Accel",
|
|
size_legend_position="none",
|
|
)
|
|
output = plot.postprocess(cars)
|
|
config = json.loads(output["plot"])
|
|
assert config["encoding"]["color"]["legend"] is None
|
|
assert config["encoding"]["shape"]["legend"] is None
|
|
assert config["encoding"]["size"]["legend"] is None
|
|
|
|
output = gr.ScatterPlot.update(
|
|
value=cars,
|
|
title="Two encodings",
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
color="Acceleration",
|
|
color_legend_position="top",
|
|
color_legend_title="Foo",
|
|
shape="Origin",
|
|
shape_legend_position="bottom",
|
|
shape_legend_title="Bar",
|
|
size="Acceleration",
|
|
size_legend_title="Accel",
|
|
size_legend_position="left",
|
|
)
|
|
|
|
config = json.loads(output["value"]["plot"])
|
|
assert config["encoding"]["color"]["legend"]["orient"] == "top"
|
|
assert config["encoding"]["shape"]["legend"]["orient"] == "bottom"
|
|
assert config["encoding"]["size"]["legend"]["orient"] == "left"
|
|
|
|
def test_update(self):
|
|
output = gr.ScatterPlot.update(value=cars, x="Horsepower", y="Miles_per_Gallon")
|
|
postprocessed = gr.ScatterPlot().postprocess(output["value"])
|
|
assert postprocessed == output["value"]
|
|
|
|
def test_update_visibility(self):
|
|
output = gr.ScatterPlot.update(visible=False)
|
|
assert not output["visible"]
|
|
assert output["value"] is gr.components._Keywords.NO_VALUE
|
|
|
|
def test_update_errors(self):
|
|
with pytest.raises(
|
|
ValueError, match="In order to update plot properties the value parameter"
|
|
):
|
|
gr.ScatterPlot.update(x="foo", y="bar")
|
|
|
|
with pytest.raises(
|
|
ValueError,
|
|
match="In order to update plot properties, the x and y axis data",
|
|
):
|
|
gr.ScatterPlot.update(value=cars, x="foo")
|
|
|
|
def test_scatterplot_accepts_fn_as_value(self):
|
|
plot = gr.ScatterPlot(
|
|
value=lambda: cars.sample(frac=0.1, replace=False),
|
|
x="Horsepower",
|
|
y="Miles_per_Gallon",
|
|
color="Origin",
|
|
)
|
|
assert isinstance(plot.value, dict)
|
|
assert isinstance(plot.value["plot"], str)
|
|
|
|
|
|
class TestLinePlot:
|
|
def test_get_config(self):
|
|
assert gr.LinePlot().get_config() == {
|
|
"caption": None,
|
|
"elem_id": None,
|
|
"interactive": None,
|
|
"label": None,
|
|
"name": "plot",
|
|
"root_url": None,
|
|
"show_label": True,
|
|
"style": {},
|
|
"value": None,
|
|
"visible": True,
|
|
}
|
|
|
|
def test_no_color(self):
|
|
plot = gr.LinePlot(
|
|
x="date",
|
|
y="price",
|
|
tooltip=["symbol", "price"],
|
|
title="Stock Performance",
|
|
x_title="Trading Day",
|
|
)
|
|
output = plot.postprocess(stocks)
|
|
assert sorted(list(output.keys())) == ["chart", "plot", "type"]
|
|
config = json.loads(output["plot"])
|
|
for layer in config["layer"]:
|
|
assert layer["mark"]["type"] in ["line", "point"]
|
|
assert layer["encoding"]["x"]["field"] == "date"
|
|
assert layer["encoding"]["x"]["title"] == "Trading Day"
|
|
assert layer["encoding"]["y"]["field"] == "price"
|
|
|
|
assert config["title"] == "Stock Performance"
|
|
assert "height" not in config
|
|
assert "width" not in config
|
|
|
|
def test_height_width(self):
|
|
plot = gr.LinePlot(x="date", y="price", height=100, width=200)
|
|
output = plot.postprocess(stocks)
|
|
assert sorted(list(output.keys())) == ["chart", "plot", "type"]
|
|
config = json.loads(output["plot"])
|
|
assert config["height"] == 100
|
|
assert config["width"] == 200
|
|
|
|
output = gr.LinePlot.update(stocks, x="date", y="price", height=100, width=200)
|
|
config = json.loads(output["value"]["plot"])
|
|
assert config["height"] == 100
|
|
assert config["width"] == 200
|
|
|
|
def test_xlim_ylim(self):
|
|
plot = gr.LinePlot(x="date", y="price", x_lim=[200, 400], y_lim=[300, 500])
|
|
output = plot.postprocess(stocks)
|
|
config = json.loads(output["plot"])
|
|
for layer in config["layer"]:
|
|
assert layer["encoding"]["x"]["scale"] == {"domain": [200, 400]}
|
|
assert layer["encoding"]["y"]["scale"] == {"domain": [300, 500]}
|
|
|
|
def test_color_encoding(self):
|
|
plot = gr.LinePlot(
|
|
x="date", y="price", tooltip="symbol", color="symbol", overlay_point=True
|
|
)
|
|
output = plot.postprocess(stocks)
|
|
config = json.loads(output["plot"])
|
|
for layer in config["layer"]:
|
|
assert layer["encoding"]["color"]["field"] == "symbol"
|
|
assert layer["encoding"]["color"]["scale"] == {
|
|
"domain": ["MSFT", "AMZN", "IBM", "GOOG", "AAPL"],
|
|
"range": [0, 1, 2, 3, 4],
|
|
}
|
|
assert layer["encoding"]["color"]["type"] == "nominal"
|
|
if layer["mark"]["type"] == "point":
|
|
assert layer["encoding"]["opacity"] == {}
|
|
|
|
def test_two_encodings(self):
|
|
output = gr.LinePlot.update(
|
|
value=stocks,
|
|
title="Two encodings",
|
|
x="date",
|
|
y="price",
|
|
color="symbol",
|
|
stroke_dash="symbol",
|
|
color_legend_title="Color",
|
|
stroke_dash_legend_title="Stroke Dash",
|
|
)
|
|
config = json.loads(output["value"]["plot"])
|
|
for layer in config["layer"]:
|
|
if layer["mark"]["type"] == "point":
|
|
assert layer["encoding"]["opacity"] == {"value": 0}
|
|
if layer["mark"]["type"] == "line":
|
|
assert layer["encoding"]["strokeDash"]["field"] == "symbol"
|
|
assert (
|
|
layer["encoding"]["strokeDash"]["legend"]["title"] == "Stroke Dash"
|
|
)
|
|
|
|
def test_legend_position(self):
|
|
plot = gr.LinePlot(
|
|
value=stocks,
|
|
title="Two encodings",
|
|
x="date",
|
|
y="price",
|
|
color="symbol",
|
|
stroke_dash="symbol",
|
|
color_legend_position="none",
|
|
stroke_dash_legend_position="none",
|
|
)
|
|
output = plot.postprocess(stocks)
|
|
config = json.loads(output["plot"])
|
|
for layer in config["layer"]:
|
|
if layer["mark"]["type"] == "point":
|
|
assert layer["encoding"]["color"]["legend"] is None
|
|
if layer["mark"]["type"] == "line":
|
|
assert layer["encoding"]["strokeDash"]["legend"] is None
|
|
assert layer["encoding"]["color"]["legend"] is None
|
|
|
|
output = gr.LinePlot.update(
|
|
value=stocks,
|
|
title="Two encodings",
|
|
x="date",
|
|
y="price",
|
|
color="symbol",
|
|
stroke_dash="symbol",
|
|
color_legend_position="top-right",
|
|
stroke_dash_legend_position="top-left",
|
|
)
|
|
|
|
config = json.loads(output["value"]["plot"])
|
|
for layer in config["layer"]:
|
|
if layer["mark"]["type"] == "point":
|
|
assert layer["encoding"]["color"]["legend"]["orient"] == "top-right"
|
|
if layer["mark"]["type"] == "line":
|
|
assert layer["encoding"]["strokeDash"]["legend"]["orient"] == "top-left"
|
|
assert layer["encoding"]["color"]["legend"]["orient"] == "top-right"
|
|
|
|
def test_update_visibility(self):
|
|
output = gr.LinePlot.update(visible=False)
|
|
assert not output["visible"]
|
|
assert output["value"] is gr.components._Keywords.NO_VALUE
|
|
|
|
def test_update_errors(self):
|
|
with pytest.raises(
|
|
ValueError, match="In order to update plot properties the value parameter"
|
|
):
|
|
gr.LinePlot.update(x="foo", y="bar")
|
|
|
|
with pytest.raises(
|
|
ValueError,
|
|
match="In order to update plot properties, the x and y axis data",
|
|
):
|
|
gr.LinePlot.update(value=stocks, x="foo")
|
|
|
|
def test_lineplot_accepts_fn_as_value(self):
|
|
plot = gr.LinePlot(
|
|
value=lambda: stocks.sample(frac=0.1, replace=False),
|
|
x="date",
|
|
y="price",
|
|
color="symbol",
|
|
)
|
|
assert isinstance(plot.value, dict)
|
|
assert isinstance(plot.value["plot"], str)
|