async def custom_aligned_metrics_streamer(requested_runs: List[AlignedRunIn],
x_axis: str) -> bytes:
for run_data in requested_runs:
run_hashname = run_data.run_id
requested_traces = run_data.traces
run = Run(hashname=run_hashname)
traces_list = []
for trace_data in requested_traces:
context = Context(trace_data.context)
trace = run.get_trace(metric_name=trace_data.metric_name,
context=context)
x_axis_trace = run.get_trace(metric_name=x_axis, context=context)
if not (trace and x_axis_trace):
continue
_slice = slice(*trace_data.slice)
iters = trace.values.sparse_numpy()[0]
sliced_iters = sliced_np_array(iters, _slice)
x_axis_iters, x_axis_values = collect_x_axis_data(
x_axis_trace, sliced_iters)
traces_list.append({
'metric_name': trace.name,
'context': trace.context.to_dict(),
'x_axis_values': x_axis_values,
'x_axis_iters': x_axis_iters,
})
run_dict = {run_hashname: traces_list}
encoded_tree = encode_tree(run_dict)
yield collect_run_streamable_data(encoded_tree)
def collect_requested_traces(run: Run,
requested_traces: List[TraceBase],
steps_num: int = 200) -> List[dict]:
processed_traces_list = []
for requested_trace in requested_traces:
metric_name = requested_trace.metric_name
context = Context(requested_trace.context)
trace = run.get_trace(metric_name=metric_name, context=context)
if not trace:
continue
iters, values = trace.values.sparse_list()
num_records = len(values)
step = (num_records // steps_num) or 1
_slice = slice(0, num_records, step)
processed_traces_list.append({
'metric_name': trace.name,
'context': trace.context.to_dict(),
'values': sliced_array(values, _slice),
'iters': sliced_array(iters, _slice),
})
return processed_traces_list