TensorFlow Backend

TensorFlow is the default backend for BaseAttentive and the most thoroughly tested path.

Installation

pip install base-attentive[tensorflow]

Or manually:

pip install tensorflow>=2.15.0

Selecting TensorFlow

import os
os.environ["KERAS_BACKEND"] = "tensorflow"

from base_attentive import BaseAttentive

model = BaseAttentive(
    static_input_dim=4,
    dynamic_input_dim=8,
    future_input_dim=6,
    output_dim=1,
    forecast_horizon=24,
)

Quick Training Example

import numpy as np
from base_attentive import BaseAttentive

model = BaseAttentive(
    static_input_dim=4,
    dynamic_input_dim=8,
    future_input_dim=6,
    output_dim=1,
    forecast_horizon=24,
    embed_dim=32,
    num_heads=4,
)

model.compile(optimizer="adam", loss="mse")

x_static  = np.random.randn(32, 4).astype("float32")
x_dynamic = np.random.randn(32, 100, 8).astype("float32")
x_future  = np.random.randn(32, 24, 6).astype("float32")
y         = np.random.randn(32, 24, 1).astype("float32")

model.fit([x_static, x_dynamic, x_future], y, epochs=3)

Accelerated Inference

Wrap repeated inference with make_fast_predict_fn to compile it with tf.function:

from base_attentive import BaseAttentive, make_fast_predict_fn
import numpy as np

model = BaseAttentive(
    static_input_dim=4, dynamic_input_dim=8, future_input_dim=6,
    output_dim=1, forecast_horizon=24,
)

x_static  = np.random.randn(32, 4).astype("float32")
x_dynamic = np.random.randn(32, 100, 8).astype("float32")
x_future  = np.random.randn(32, 24, 6).astype("float32")

fast_predict = make_fast_predict_fn(
    model,
    warmup_inputs=[x_static, x_dynamic, x_future],
)
predictions = fast_predict([x_static, x_dynamic, x_future])

Note

Keep batch and sequence shapes stable across calls for best results. For training, model.compile(..., jit_compile="auto") may also help.

Compatibility Check

from base_attentive.backend import check_tensorflow_compatibility

ok, msg = check_tensorflow_compatibility()
print(msg)

Minimum required version: TensorFlow 2.15.0.