Overview

In addition to the 5,400+ catalog blocks loaded from the CSV, BCILattice injects a set of custom built-in blocks at server startup. These cover analysis utilities, pipeline I/O, code execution, file selection, and Workflow graph controls.

New here? The Block Index lists every one of our blocks in a single line, grouped as the palette groups them.

Custom blocks are not classes; they run as functions or server-side routines and appear in the block palette alongside catalog blocks. They are marked is_custom: true internally.

Analysis Blocks

These blocks call the local training server's analysis routes (/xlb9/…) and are available in both the ML Suite and Workflow Designer canvases.

Channel Selection

Ranks and filters EEG/fNIRS/EMG channels by statistical relevance. Calls /xlb9/channel_selection/run. The Workflow canvas exposes an extra data_type selector; the ML Suite version omits it.

Settings

Custom Labels

Injects or overrides the label column used by downstream blocks. Accepts a path to a label CSV or a comma-separated string of manual labels.

Settings

Signal Projection

Converts 2-D time-series signals into 2-D images using time-frequency or recurrence transforms. Calls /xlb9/signal_projection/run. Useful for feeding EEG into CNN-based image classifiers.

Settings

Time-Frequency Features

Turns the Analysis suite's time-frequency output into a trainable feature table. It reduces a per-trial ERSP cube — (epochs, channels, freqs, times) — down to X[n_trials, n_features] with labels that stay index-aligned, so a time-frequency analysis can actually train a decoder instead of only rendering a figure.

Settings

Pipeline Blocks

Structural blocks that control data flow and code execution within a pipeline. Available in both canvases unless noted otherwise.

Input Block

Marks one entry point of the pipeline, and nothing else. Drag one onto the canvas for each data stream the pipeline accepts. Each Input Block becomes exactly one input port on the compiled-pipeline block when this pipeline is dropped into a Workflow, which tells you where to connect what. Which files arrive, how they split into runs and how they are filtered are set on the Workflow's Select Files.

Settings

File Path

Points at one file on disk and emits its path. It reads nothing — that is the point: a block that does its own file reading needs the path, and Select Files cannot give it one usefully, because Select Files exists to load a processing stage into a table.

Settings

Output Block

Marks one exit point of the pipeline. Whatever is connected here becomes one output port on the compiled-pipeline block. Every pipeline needs exactly one final output node — an Output Block, Graph Output or Graph Endpoint — or compiling fails.

Settings

Loop Block

Repeats a sub-section of the pipeline for N iterations or epoch-by-epoch. Set start_node and end_node to the labels of the blocks that mark the loop boundary.

Settings

Custom Code

Executes an arbitrary Python function inline inside the pipeline. The function receives the upstream data tensor as its first argument and must return the transformed result.

def custom_process(data):
    # data is the input tensor from the previous block
    # transform it here
    return data

Settings

Model Structure Blocks

The only two blocks that take models in and hand a model out. Everything else on the canvas moves data; these build the shape of one network.

Chaining layer blocks already composes them in order — these are for the case a chain cannot express: several branches over the same input, joined back together. Each wired Branch N receives the identical input inside the composed module's forward.

Model Concat

Runs every wired branch on the same input and concatenates the results into one tensor. This is how a multi-scale CNN or a two-stream model is drawn: three Conv1d branches with different kernel sizes, joined here, then a shared head.

Settings

Model Add

Runs every wired branch on the same input and sums the results. Put a transformed path on one branch and an Identity layer on the other, and you have a residual/skip connection.

Settings

Trainer Trigger

Hands one side of the trainer's train/test split to a block that fits on its own. Use it when a model block shows X and y ports — that block is a constructor and a fit call in one, and it must not see the rows the trainer will score against.

Settings

Graph Blocks

Four ways to get a picture out of a run. The important distinction is tap vs. terminal: Data Graph passes its value straight through, so it can be spliced anywhere, any number of times. Graph Output and Graph Endpoint are terminal nodes, so they compete for the “exactly one final output” slot every compiled graph must have.

Data Graph

A general-purpose plot tap: it renders whatever reaches it and passes the same value out unchanged on its Data port. Because it is not a terminal node, you can drop it onto an existing connection to inspect what is flowing through mid-pipeline without rewiring anything downstream.

Settings

Confusion Matrix Graph

Draws a confusion matrix from true and predicted labels. Wire it to a trainer's Test Labels and Test Predictions outputs (out-of-fold when a Validation Scheme is wired).

Settings

Graph Output

Terminal plot node: renders the incoming value and saves it to disk. Counts as a pipeline's single final output, so it can end a graph in place of an Output Block.

Settings

Graph Endpoint

The other terminal plot node. Behaves like Graph Output but writes to its own output_path, and is the node the workflow HTTP route targets when a graph is executed over REST.

Settings

Workflow-Only Blocks

Loading real recordings is a Workflow-canvas concern, so this block exists only there. An ML Suite pipeline takes its data through an Input Block instead, and is fed by whatever the Workflow wires into it.

Select Files

Loads real recordings from the Data Manager: pick a modality, a processing stage, a session and the subjects, and it emits the feature table X and labels Y. This is where a workflow actually gets its data.

Settings

Training several models on what Select Files loads? A trainer's Model input takes one wire — the canvas refuses a second one and offers to insert a Model Collection instead, so every subject's recordings load once and each model gets its own results row. See Model Collection in the Advanced Blocks reference.

Compiled Pipeline Block

When you compile and register a pipeline from the ML Suite or Workflow Designer, BCILattice dynamically injects it as a reusable block into the Pipeline Blocks category. These blocks are created at runtime via POST /xlb9/register_compiled_pipeline or POST /xlb9/register_compiled_workflow.

Compiled Pipeline

A compiled and registered pipeline exposed as a single reusable block, collected in the Model/Pipelines (Compiled) palette group. It has one input port per Input Block and one output port per Output Block in the original pipeline, each named from that block's port_name.

Settings