Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.atomscale.ai/llms.txt

Use this file to discover all available pages before exploring further.

The RHEEDStreamer class lets you push RHEED frames directly from your instrument to Atomscale for live analysis. Choose between two streaming modes:
  • Callback / push mode - The camera or SDK hands you fresh frames and you upload each chunk immediately
  • Generator / pull mode - You already have frames buffered and want the helper to pace the upload for you

​
Prerequisites

  • numpy installed
  • RHEED frames as uint8 arrays shaped (N, H, W) or (H, W)
  • A stable clock so you can honor the capture cadence

​
Create a streamer

from atomscale.streaming import RHEEDStreamer

streamer = RHEEDStreamer(api_key="YOUR_API_KEY")
Optional keyword arguments tune chunking and logging. Set verbosity=4 for detailed progress output.

​
Callback / push mode

Use this variant when frames arrive live from the instrument. The outer loop is your acquisition callback: once a chunk is ready, send it to the API and wait just long enough to match the capture cadence. 
import numpy as np
import time

fps = 120.0
chunk_size = 240  # 2 seconds of frames at 120 fps
seconds_per_chunk = chunk_size / fps

data_id = streamer.initialize(
    fps=fps,
    rotations_per_min=15.0,  # set to 0.0 for stationary
    chunk_size=chunk_size,
    stream_name="Demo (callback mode)",
    physical_sample="Demo wafer",
)

for chunk_idx in range(5):
    # Replace with actual frame capture
    frames = np.random.randint(0, 256, size=(chunk_size, 300, 500), dtype=np.uint8)
    streamer.push(data_id, chunk_idx, frames)
    time.sleep(seconds_per_chunk)

time.sleep(1.0)  # let in-flight uploads finish
streamer.finalize(data_id)

​
Generator / pull mode

Use this form when frames are already buffered (from disk, memory, or a simulated source). Provide an iterator that yields chunks and the helper handles pacing and retry logic. 
import numpy as np
import time

def frame_chunks(frames, *, chunk_size=240, fps=120.0):
    seconds_per_chunk = chunk_size / fps
    for start in range(0, len(frames), chunk_size):
        yield frames[start : start + chunk_size]
        time.sleep(seconds_per_chunk)

# Simulated buffered frames
frames = np.random.randint(0, 256, size=(1200, 300, 500), dtype=np.uint8)

data_id = streamer.initialize(
    fps=10.0,
    rotations_per_min=0.0,
    chunk_size=20,
    stream_name="Demo (generator mode)",
    physical_sample="Demo wafer",
)

streamer.run(data_id, frame_chunks(frames, chunk_size=20, fps=10.0))
streamer.finalize(data_id)

​
Best practices

RecommendationWhy
Maintain original capture cadenceThe server expects frames at the declared FPS
Use chunk sizes of at least 2 secondsReduces overhead and improves throughput
Always call finalize()Lets the pipeline clean up gracefully, even on failure
Use distinct stream namesMakes it easier to find runs later
Always call finalize() even if the upload fails part-way. This signals the server to clean up resources and mark the stream appropriately.
Pass physical_sample to initialize() to link the stream to an existing sample. Names are matched case-insensitively, or a new sample is created if no match is found. You can also pass project_id to associate the stream with a project. 
data_id = streamer.initialize(
    fps=120.0,
    rotations_per_min=15.0,
    chunk_size=240,
    stream_name="Growth run 42",
    physical_sample="GaN-001",
    project_id="your-project-uuid",
)

​
Next steps

Poll Time Series

Monitor streaming analysis in real time.

Streaming Reference

Full RHEEDStreamer API documentation.