geocene / trainset Goto Github PK

It's pretty hard to click and individual point - sometimes it seems like I am hovering over the right spot, but then I click and it doesn't work. Just because my curser changes from the crosshairs to the pointer, it doesn't mean that when I click I will hit the point under my pointer.

SELECTION
CLICK: toggle label for a point
CLICK & DRAG: apply TRUE label to selection
+SHIFT: apply FALSE label
+OPTION: centered selection
+CONTROL: full Y selection

NAVIGATON
→ or ← : pan
SHIFT + → or ← : fast pan
↑ or ↓: zoom
CLICK bottom context bar: unfocus
CLICK & DRAG bottom context bar: adjust focus region

What happens when someone uploads a CSV where there are multiple values in the filename column. Do we allow this? Is there a rational use case for the CSV to be named something other than what's in the filename column? Do we want to support some case where someone uploads a CSV called all_my_files.csv that has like 20 different filenames (I guess with non-overlapping timestamps?) in it?

Danny's option: the CSV's filename has the match the filename column, and all values in the filename column have to be the same.

The circle on the max and min values on the context window get chopped off

https://help.heroku.com/J2R1S4T8/can-heroku-force-an-application-to-use-ssl-tls

Likely because line 84 in index.vue doesn't take into account renaming of files after they have been labeled. Commenting out this line as a temporary fix works.

Let's try to prevent this sort of thing:

space moves focus window to the right 90% of the window width
shift+space moves the focus window to the left 90% of the window width

Right now the chart resizing seems like more of a liability than a feature. If it takes less than 1 hour, let's fix it. If it take more than 1 hour, let's scrap it.

Make this available for download immediately below the "Upload" button and linked in the help docs.
sample_trainset.csv.zip

I should be able to easily see which file I am labeling while labeling - if I send a screenshot, the person I send it to should be able to know which file this is from.

sample_trainset.csv.zip

Basically, if I hold CONTROL, make it work the way the context bar works. SHIFT+CONTROL should deselect everything in the x region I brushed

If I drag the focus slider (is that what it's called?) to make it a lot wider, half the points in the focus window are behind the line and half are in front of the line:

Just after pressing "Clear" -- looks as expected

Scroll focus window one half screen to the right, and now we see some uncleared labels

The style is a little wonky, could use some better branding, spinners need help, and generally it just needs a little TLC. Geocene's style guide is here.

I partially labeled a 32k datapoint file, downloaded it, and tried to re-upload to continue labeling. Resulted in "Upload Failed | Make sure data is in the TRAINSET format. See help" error.

Make this a straight line instead:

Failed - network error

Add some padding around the left, right, and top of the chart - it looks nicer! Here's a mockup:

Default to TRUE brush, shift key held makes brush FALSE, get rid of INVERT brush

TRAINSET will hopefully be used by a lot of non-cookstove people. Name the brushes:

invert
label true
label false

Right now I press export->upload->home page upload->finder->label, but I shouldn't have to go back to the homepage. When I press this upload button, I should be able to upload the new file:

Make the color that purple instead of the red, and give it the same opacity as the black points

To reproduce:

• click sample CSV
• you will get a broken page
• press back
• the help page looks like this:

Have a box in the top right area that shows me the timestamp and value of the dot that my cursor is hovering over

Right move focus box right by 20%
Left move focus box left by 20%
Up zoom in 20%
Down zoom out 20%

Figure out a way to make it easy to make updates after it's deployed

Let's remove this wrapping feature, or at least only make it show up when the window gets very small:

If values in exported files are Null, import fails. Here's a sample file. How do we want to deal with this?

LPG_PK_01_TC_9805_LPGL_20171201_177.lablr.csv.zip

Right now it looks like the y axis goes exactly from data_min to data_max. This can make it really hard to click or brush points near the extremes. Let's draw the y-axis this way: y_max = data_max + (data_max - data_min) * 0.05 and y_min = data_min - (data_max - data_min) * 0.05

I cannot get any files to upload on fa3d6d0

All date strings are exported as UTC standard

Suffix is currently -lablr. Change it to -labeled.

On bigger files, like the 32k files, we can set the focus width (the selection in the context) to the entire width. This slows things down. Can we limit the width to a certain number of points?

Options:
something like this: crossfilter.github.io/crossfilter/ (geocene.github.io/trainset?)
or: trainset.availabledomain
or: trainset.geocene.com
or: trainset.sumsarizer.com
or: deploy it yourself, bum! (Danny doesn't like this option)

Is one of the format rules that the dataset needs to have fewer rows than a certain threshold, or is this 3E5 dummy_input dataset improperly formatted?