Streaming TT (STTA)

One-pass randomised TT approximation for data too large to materialise fully. STTA processes data in a streaming fashion, building the TT representation incrementally.

Basic Usage

from tinytt.streaming import StreamingTT, streaming_tt

# Generate some data (simulating a stream)
import numpy as np
data = np.random.randn(100, 4, 4, 4).astype(np.float64)

# Convenience function (loads all data into memory internally)
x_tt = streaming_tt(shape=[4, 4, 4], ranks=[3, 3], data=data)
print(x_tt.R)          # [1, 3, 3, 1]

Streaming Object (Incremental)

For true streaming workloads where data arrives incrementally:

stream = StreamingTT(shape=[4, 4, 4], ranks=[3, 3])

# Feed data slices one by one
for batch in data_generator():
    stream.insert(batch)

# Retrieve final TT
x_tt = stream.tt()

Data Sources

The data parameter accepts:

• A tensor (sliced along dim 0)
• A callable returning an iterator
• Any iterable of tensor slices

# Callable source
def data_stream():
    for _ in range(100):
        yield np.random.randn(4, 4, 4)

x_tt = streaming_tt(shape=[4, 4, 4], ranks=[3, 3], data=data_stream)

Curvature-Aware Streaming

For problems where the intrinsic curvature matters:

from tinytt.streaming import StreamingCurvature

curv = StreamingCurvature(shape=[4, 4, 4], ranks=[3, 3])
curv.insert(data_batch)
x_tt = curv.tt()

When to Use Streaming TT

• Large datasets that don't fit in memory
• Online learning where new data arrives over time
• Single-pass constraints where revisiting data is expensive

Further Reading

• Tests: PYTHONPATH=. pytest tests/test_streaming.py tests/test_streaming_convergence.py -v
• streaming.py