Repository navigation
feature: add lean ui terminal UI with live backtest results - #675
Open
jaredbroad wants to merge 15 commits into
Open
jaredbroad wants to merge 15 commits into
jaredbroad wants to merge 15 commits into
Conversation
…commands Adds a keyboard-driven terminal UI built on Textual. It lists local projects under the Lean CLI root and, when logged in, the organization's cloud projects, labelled local, cloud or synced. Projects can be filtered and navigated with the arrow keys or j/k. The selected project offers backtest, cloud backtest, logs, push, pull and cloud status actions. Each one shows the exact `lean` command for confirmation, runs it as a child process (`python -m lean`) and streams its output into the UI. `x` stops a running command via Ctrl+C (Ctrl+Break on Windows), and pressing it again kills the process. Textual is only imported by `lean ui`, so other commands start as fast as before. textual-plotext is added for the upcoming backtest charts. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Pressing `v` on a project with local backtests opens a results screen modeled on the QuantConnect.com backtest page: a list of the project's backtests, the runtime statistics, equity against a scaled benchmark, drawdown, and tabs for statistics, orders, every other chart (series in a second unit go on a right-hand axis) and the log. Results are parsed from the result file LEAN writes, which uses the same chart, order and statistics format as the cloud API, so the parser can serve cloud backtests too. Running backtests report progress from their last equity point. Chart axes use round values and round times. Also fixes the projects screen being queried while another screen was on top, which failed when cloud projects finished loading behind a dialog, and stops project action keys from firing on other screens. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…testing The results screen now shows a project's cloud backtests next to its local ones, newest first, read through the API: the backtest summary and its Strategy Equity chart, with orders (100 per request) and logs (200 per request) loaded when their tab is opened. Backtests that are still running are refreshed every 15 seconds in the cloud and every 10 seconds locally. `b` and `c` now open the results right away instead of asking to confirm. A local backtest is written to an output directory the UI chooses, so its results can be read while it runs; its progress comes from the date range in LEAN's log, since the result file only gets it at the end. A cloud backtest switches to the backtest as soon as `lean cloud backtest` prints its url. The command's output is the backtest's log while it runs. The results screen uses the full width: `[` and `]` step through backtests and `o` opens a filterable list of all of them. Only the Strategy Equity chart is shown. The latest logs action is replaced by the Logs tab. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The x axis now runs from the backtest's start to its end date when they are known, so a running backtest's equity grows from left to right instead of being stretched over the full width. Local backtests take the period from LEAN's log while running and cloud backtests from the API. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The orders tab loaded every order before showing any, which for a cloud
backtest with millions of orders meant tens of thousands of requests. It now
loads 100 orders when the tab opens and the next 100 as the cursor or view
nears the end, showing "Orders (100 of 3,555)". While a backtest runs only
orders placed since the last refresh are requested, and the orders are
loaded again once it completes so their final status shows.
The cloud chart, orders and log endpoints answer {"status": "loading"}
while they prepare a backtest's data, which left the orders tab empty after
a backtest completed. Those answers are now retried, and if the statistics
count orders the endpoint does not return yet, loading them is retried too.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The test added the orders while retries ran every 0.1 seconds, and waiting for the app's workers could last through all of them, so it sometimes failed. The fake endpoint now returns the orders from its third request. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The cloud orders endpoint answers a running backtest with no orders and a length of 100, which showed as "Orders (0 of 100)". The length is now only used when a page has orders, and a running backtest without any shows "Orders (none yet)". Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A completed and a running cloud backtest of the Intraday HMM Simplified Regime Detection project, for the pull request description. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Member
Author
|
🍿 #vibed |
- Run commands with the lean-cli binary in the portable build - Elevate `lean ui` with sudo on Linux so its commands need no terminal - Claim the running command on the UI thread so only one runs at a time - Treat null order fields as missing instead of crashing the orders tab - Keep the selected backtest when a cloud backtest gets its id - Store command output before the results screen is mounted - Show command output in batches instead of a round trip per line - Read cloud backtests through BacktestClient and QCBacktest - Share the stored-credentials check with the whoami message - Scope imports to the functions that use them
- Keep the new QCBacktest fields lenient, the API sends [] and "" for empty values - Wait for the running command to stop before quitting instead of hanging on exit - Mark the commands `lean ui` runs with LEAN_UI_COMMAND so cancelling a cloud run needs no prompt and Ctrl+Break only acts as Ctrl+C in them - Honor a stop asked before the command has started - Reuse QCLanguage for the project language labels - Mark the live log loaded when a cloud backtest gets its id - Evaluate each action's availability once per shown project
- Interrupt a command once per stop, force stop its containers on the second and kill it on the third - Do not escalate a stop already asked for when quitting - Stop containers gracefully on the first interrupt of a command `lean ui` runs - Reuse QCBacktest.is_complete to tell when a cloud backtest has finished - Read the output directory config through OutputConfigManager - Name the output directory format once and use it everywhere - Evaluate the `lean ui` command flag once in constants
- Scope the projects screen handlers to its own table and action list - Stop polling a backtest whose result failed to load, ask to press r instead - Treat a result file that cannot be read like one still being written - Ask whether to delete a running cloud backtest when stopping it, instead of auto-confirming the cloud runner prompt - Import single-use standard library names where they are used - Move the `lean.ui` tests under tests/ui
- Cover the project table with a spinner until the first projects arrive - Show "loading projects..." in the table subtitle from the start - Focus the table once it has loaded, a loading widget cannot take focus
pyfakefs switches to the real filesystem process-wide for short moments, so Textual's worker threads sometimes read the real disk instead of the fake one. The two tests that build a screen outside the app now do so in an event loop, which Python 3.9 needs for the lock every widget creates.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds
lean ui, a keyboard-driven terminal UI (in the style of btop) for browsing projects, running backtests and watching their results, local or in the cloud.Completed cloud backtest of the Intraday HMM Simplified Regime Detection project:
While a backtest runs, the equity chart fills in from left to right over the backtest's period and refreshes every 15 seconds:
Running cloud backtest at 42%:
What it does
local,cloudorsynced.↑↓/j kto move,/to filter,enterfor actions.b/cbacktest locally / in the cloud and open the results right away. A local backtest is written to an output directory the UI picks so it can be read while running; a cloud backtest switches over as soon aslean cloud backtestprints its url. The command's output is the Logs tab while it runs.vview results of every local and cloud backtest of the project, newest first.[]step through them,oopens a filterable list.Orders (100 of 3,555)), so backtests with millions of orders stay fast.xstops a running command the way Ctrl+C would.How it works
lean uiruns CLI commands as child processes (python -m lean), so their console output can't draw over the UI and no command needed refactoring.lean/ui/results.pyreads both. In the cloud only the Strategy Equity chart is read, with orders (100 per request) and logs (200 per request) loaded when their tab opens.{"status": "loading"}while they prepare a backtest's data, and a running backtest's orders come back empty with a length of 100. Both are handled.lean ui, so other commands start as before.textualandtextual-plotextare added toinstall_requires(about 1.9 MB of wheels).Testing
tests/commands/test_ui.py), the result readers (tests/test_ui_results.py) and axis ticks (tests/test_ui_ticks.py). Static analysis passes.cfrom the UI end to end.🤖 Generated with Claude Code