Create, visualize and deploy AI solutions.

The pipeline steps should be collapsable to maximize space. When clicking a selected block, it should deselect (implemented) and also collapse the pipeline and show only an "expand" icon like "v". Ideally there should be an animation when the pipeline collapses to indicate the pipeline is still available.

closes https://github.com/hal9ai/hal9/issues/162

@javierluraschi we now return something like

basically for each function response we always have a result enum that can either be Value (arbitrary json, like before) or Error (string with error message)

In order to create and test some examples under say /examples/controls-basics we need a way to start the server in that folder with a custom .R or .py file, so adding h9_start() for this. It takes an optional server parameter with the server file to use.

The current default link is a test file inside a private git repository. Would recommend moving the link to a consistent location like the other import(json,csv...)

https://medium.com/@dharini_r/5fdceb7b09c7

importing hal9 gives an error — import controls — module not found stating the path “/home/username/anaconda3/envs/hal9_project/lib/python3.9/site-packages/hal9.py”

Does not work:

library(hal9)

data_path <- 'https://temphal9.s3.us-west-2.amazonaws.com/brenda/penguins.csv'

penguins <- read.csv(data_path)

h9_dropdown(
    "X_dropdown",
    values = function() {
      colnames(penguins)
    },
    on_update = function(value) {
        h9_set(value, "X_val")
    }
)

does work:

library(hal9)


h9_dropdown(
    "X_dropdown",
    values = function() {
      data_path <- 'https://temphal9.s3.us-west-2.amazonaws.com/brenda/penguins.csv'

      penguins <- read.csv(data_path)

      colnames(penguins)
    },
    on_update = function(value) {
        h9_set(value, "X_val")
    }
)

Looks like bad R code breaks the server... notice typo h9_images instead of h9_image:

h9_markdown("md", "This is a test!!!")

h9_images("image", "https://cdn4.buysellads.net/uu/1/119121/1658931414-carbon-pubdev_2x.png")

from @javierluraschi : hit a case where hot reload stopped working... I think it might be after I save an invalid R file (like function typo functttion()) that the R session is willied and hot-reload doesn't work anymore (I think).

@javierluraschi @anchit-sadana realized we could simplify the json specs a bit, lmk if objection

proposed:

{
    "manifests": [
        {
            "runtime": "r",
            "calls": [
                {
                    "node": "species_dropdown",
                    "fn_name": "on_select",
                    "args": [
                        {
                            "name": "value",
                            "value": "setosa"
                        }
                    ]
                },
                {
                    "node": "summary_table",
                    "fn_name": "html",
                    "args": []
                }
            ]
        }
    ]
}

old:

{
    "manifests": [
        {
            "runtime": "r",
            "nodes": [
                {
                    "name": "species_dropdown",
                    "calls": [
                        {
                            "fn_name": "on_select",
                            "args": [
                                {
                                    "name": "value",
                                    "value": "setosa"
                                }
                            ]
                        }
                    ]
                },
                {
                    "name": "summary_table",
                    "calls": [
                        {
                            "fn_name": "html",
                            "args": []
                        }
                    ]
                }
            ]
        }
    ]
}

@anchit-sadana hopefully fastapi already gives us this but we need a way to determine the port on which the service is listening on, in the case where the user doesn't provide the port

Mostly, bug fixes where performed in the runtime / designer already. Just adding support for ?localhost&debug to activate the script window to help us add breakpoints for the JS controls. Working on a more complex app with all that currently looks as follows:

I'm planning to add all the controls in a single app to showcase them as a example and help us build more complex apps.

1. Add controls
2. Use control + click to select multiple elements

Problem: Controls remain grouped even after clicking each of them and not using control click.

The following code does not change the value of the html element since the client is not resetting state:

import hal9 as h9

def updated(value):
  h9.set("length", value.length)

h9.textarea("textarea", on_update=updated)

h9.html("html", rawhtml=lambda: h9.get("length"))

Crashes while processing text files bigger than 1mb.

import hal9 as h9

h9.textarea("textarea", on_update=lambda value: h9.set("textarea", value))

def onclick():
  big_string = h9.get("textarea")
  h9.html("html2", rawhtml=lambda: some_function(big_string))

h9.node("button", on_click= onclick, label=lambda: 'Process text', kind=lambda: 'primary')

Reprex:

h9 --platform python new app1
h9 start app1

Notice that by default /design path will be open, remove /design to navigate to /. This is the runtime mode (not design of the application), notice that the app does not apply the layout.