Delete text2roi_colab_demo.ipynb

text2roi_colab_demo.ipynb

@@ -1,301 +0,0 @@
-{
-  "cells": [
-    {
-      "cell_type": "markdown",
-      "metadata": {},
-      "source": [
-        "# Amphora NeuroText — Colab Demo\n",
-        "\n",
-        "**Type text → predict 56 brain ROI activations → visualize on fsaverage5 cortical maps (MNE + Nilearn).**\n",
-        "\n",
-        "Open this notebook via the **GitHub Colab link** on the model card (recommended).\n",
-        "\n",
-        "Weights download from the public HF repo `ffh92r32rm0/Amphora_NeuroText` — no login required.\n",
-        "\n",
-        "Pipeline:\n",
-        "```\n",
-        "text → Qwen3-Embedding-4B → Text2ROI MLP (3M params) → 56 ROI scores → HCP-MMP parcels → fsaverage5 surface\n",
-        "```\n",
-        "\n",
-        "Brain plotting uses **MNE + Nilearn only** (BSD/MIT-friendly) — no tribev2 import (CC BY-NC).\n",
-        "\n",
-        "**Setup:** Runtime → Change runtime type → **GPU** (T4 is enough), then run all cells.\n",
-        "\n",
-        "Model: [ffh92r32rm0/Amphora_NeuroText](https://huggingface.co/ffh92r32rm0/Amphora_NeuroText) · MIT license\n",
-        "\n",
-        "> ⚠️ Predictions are **model estimates** from group fMRI training — not measurements of anyone's private thoughts."
-      ]
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "# Install dependencies (first run ~5–8 min)\n",
-        "# Commercial-friendly: MNE + Nilearn for brain viz — no tribev2 (CC BY-NC).\n",
-        "import sys\n",
-        "\n",
-        "!{sys.executable} -m pip install -q --upgrade pip\n",
-        "!{sys.executable} -m pip install -q \"torch>=2.3\" \"transformers>=4.44\" huggingface_hub ipywidgets\n",
-        "!{sys.executable} -m pip install -q nilearn mne nibabel scipy matplotlib numpy\n",
-        "\n",
-        "print(\"Deps installed (MNE + Nilearn for fsaverage5 brain maps).\")"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "import json\n",
-        "from pathlib import Path\n",
-        "\n",
-        "import matplotlib.pyplot as plt\n",
-        "import numpy as np\n",
-        "import torch\n",
-        "import torch.nn as nn\n",
-        "import torch.nn.functional as F\n",
-        "from huggingface_hub import hf_hub_download\n",
-        "from IPython.display import display, clear_output\n",
-        "import ipywidgets as widgets\n",
-        "\n",
-        "HF_REPO = \"ffh92r32rm0/Amphora_NeuroText\"\n",
-        "\n",
-        "DEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
-        "print(f\"Device: {DEVICE}\")\n",
-        "\n",
-        "CKPT_PATH = hf_hub_download(HF_REPO, \"text2roi_projector.pt\")\n",
-        "print(f\"Checkpoint: {CKPT_PATH}\")"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "ROI_NAMES_56 = [\n",
-        "    \"V1\", \"V2\", \"V3\", \"V4\", \"V3A\", \"V3B\", \"LO1\", \"LO2\",\n",
-        "    \"MT\", \"MST\", \"V7\", \"IPS1\",\n",
-        "    \"FFA-1\", \"FFA-2\", \"PPA\", \"RSC\", \"OFA\", \"EBA\",\n",
-        "    \"IPS2\", \"IPS3\", \"IPS4\", \"IPS5\", \"SPL1\", \"hIP1\", \"hIP2\", \"hIP3\",\n",
-        "    \"dlPFC\", \"vlPFC\", \"OFC\", \"ACC\", \"mPFC\", \"FP1\", \"FP2\",\n",
-        "    \"IFG\", \"IFGorb\", \"STG\", \"STS\", \"MTG\", \"AG\",\n",
-        "    \"PCC\", \"mPFC_dmn\", \"LP_L\", \"LP_R\", \"HPC_L\", \"HPC_R\",\n",
-        "    \"AI\", \"dACC\", \"sgACC\", \"vmPFC\",\n",
-        "    \"Amygdala_L\", \"Amygdala_R\", \"Caudate_L\", \"Caudate_R\",\n",
-        "    \"Putamen_L\", \"Putamen_R\", \"Thalamus\",\n",
-        "]\n",
-        "\n",
-        "class Text2ROI(nn.Module):\n",
-        "    def __init__(self, in_dim=2560, hidden=1024, out_dim=56, dropout=0.1):\n",
-        "        super().__init__()\n",
-        "        self.net = nn.Sequential(\n",
-        "            nn.Linear(in_dim, hidden), nn.GELU(), nn.Dropout(dropout), nn.LayerNorm(hidden),\n",
-        "            nn.Linear(hidden, hidden // 2), nn.GELU(), nn.Dropout(dropout),\n",
-        "            nn.Linear(hidden // 2, out_dim),\n",
-        "        )\n",
-        "    def forward(self, x):\n",
-        "        return self.net(x)\n",
-        "\n",
-        "def load_projector(path):\n",
-        "    ckpt = torch.load(path, map_location=\"cpu\", weights_only=False)\n",
-        "    model = Text2ROI(int(ckpt[\"in_dim\"]), int(ckpt[\"hidden\"]), int(ckpt[\"n_roi\"]))\n",
-        "    model.load_state_dict(ckpt[\"state_dict\"])\n",
-        "    model.to(DEVICE).eval()\n",
-        "    names = [str(x) for x in ckpt.get(\"roi_names\", ROI_NAMES_56)]\n",
-        "    return model, names, ckpt\n",
-        "\n",
-        "PROJECTOR, ROI_NAMES, CKPT_META = load_projector(CKPT_PATH)\n",
-        "print(f\"Loaded Text2ROI v1 — val R={CKPT_META.get('best_val_r', 0):+.3f}, {len(ROI_NAMES)} ROIs\")"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "QWEN_ID = \"Qwen/Qwen3-Embedding-4B\"\n",
-        "from transformers import AutoModel, AutoTokenizer\n",
-        "\n",
-        "print(\"Loading Qwen3 embedder (first run downloads ~8 GB)...\")\n",
-        "_tok = AutoTokenizer.from_pretrained(QWEN_ID, padding_side=\"left\")\n",
-        "_qwen = AutoModel.from_pretrained(QWEN_ID, dtype=torch.bfloat16).to(DEVICE).eval()\n",
-        "print(\"Qwen3 ready.\")\n",
-        "\n",
-        "@torch.no_grad()\n",
-        "def embed_text(text: str) -> np.ndarray:\n",
-        "    tok = _tok([text], return_tensors=\"pt\", padding=True, truncation=True, max_length=512).to(DEVICE)\n",
-        "    out = _qwen(**tok)\n",
-        "    emb = out.last_hidden_state[:, -1].float()\n",
-        "    emb = F.normalize(emb, p=2, dim=1).cpu().numpy().astype(np.float32)\n",
-        "    return emb\n",
-        "\n",
-        "@torch.no_grad()\n",
-        "def predict_rois(text: str) -> dict:\n",
-        "    x = torch.from_numpy(embed_text(text)).to(DEVICE)\n",
-        "    pred = PROJECTOR(x).cpu().numpy().reshape(-1)\n",
-        "    return {name: float(pred[i]) for i, name in enumerate(ROI_NAMES)}"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "# Map 56 ROI values → fsaverage5 vertices (MNE + Nilearn — no tribev2)\n",
-        "import sys\n",
-        "import urllib.request\n",
-        "from pathlib import Path\n",
-        "\n",
-        "_helper = Path(\"text2roi_to_fsaverage.py\")\n",
-        "if not _helper.exists():\n",
-        "    try:\n",
-        "        from huggingface_hub import hf_hub_download\n",
-        "        _helper = Path(hf_hub_download(HF_REPO, \"text2roi_to_fsaverage.py\"))\n",
-        "    except Exception:\n",
-        "        urllib.request.urlretrieve(\n",
-        "            \"https://raw.githubusercontent.com/hamcoderfran/neuroevolution/main/huggingface/text2roi/text2roi_to_fsaverage.py\",\n",
-        "            _helper,\n",
-        "        )\n",
-        "sys.path.insert(0, str(_helper.parent))\n",
-        "from text2roi_to_fsaverage import roi_dict_to_fsaverage5, plot_fsaverage5_brain\n",
-        "\n",
-        "print(\"fsaverage5 plotting helpers ready (MNE + Nilearn).\")"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "def plot_results(text: str, roi_values: dict):\n",
-        "    vertex_map, skipped = roi_dict_to_fsaverage5(roi_values)\n",
-        "\n",
-        "    # --- ROI bar chart ---\n",
-        "    top = sorted(roi_values.items(), key=lambda kv: abs(kv[1]), reverse=True)[:16]\n",
-        "    names = [t[0] for t in top]\n",
-        "    vals = [t[1] for t in top]\n",
-        "    colors = [\"#c44e52\" if v < 0 else \"#4c72b0\" for v in vals]\n",
-        "\n",
-        "    fig, ax = plt.subplots(figsize=(7, 5))\n",
-        "    ax.barh(names[::-1], vals[::-1], color=colors[::-1])\n",
-        "    ax.axvline(0, color=\"#333\", lw=0.8)\n",
-        "    ax.set_title(\"Top ROI predictions\")\n",
-        "    ax.set_xlabel(\"Predicted activation\")\n",
-        "    plt.tight_layout()\n",
-        "    plt.show()\n",
-        "\n",
-        "    # --- fsaverage5 brain maps (Nilearn) ---\n",
-        "    skip_note = f\" — subcortical skipped: {len(skipped)}\" if skipped else \"\"\n",
-        "    plot_fsaverage5_brain(\n",
-        "        vertex_map,\n",
-        "        views=[\"left\", \"right\", \"dorsal\"],\n",
-        "        cmap=\"RdBu_r\",\n",
-        "        threshold=0.02,\n",
-        "        title=f\"fsaverage5 cortical map (HCP-MMP ROIs){skip_note}\",\n",
-        "    )\n",
-        "    plt.show()\n",
-        "\n",
-        "    print(\"\\nTop ROIs:\")\n",
-        "    for n, v in top[:10]:\n",
-        "        print(f\"  {n:<16} {v:+.4f}\")\n",
-        "    if skipped:\n",
-        "        print(f\"\\n(Subcortical / unmapped ROIs not painted on cortex: {', '.join(skipped[:8])}{'…' if len(skipped)>8 else ''})\")"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {},
-      "source": [
-        "## Try it — enter text and click **Predict**\n",
-        "\n",
-        "Or use a preset example below."
-      ]
-    },
-    {
-      "cell_type": "code",
-      "metadata": {},
-      "source": [
-        "text_in = widgets.Textarea(\n",
-        "    value=\"a dog running through a sunny park\",\n",
-        "    description=\"Text:\",\n",
-        "    layout=widgets.Layout(width=\"90%\", height=\"80px\"),\n",
-        ")\n",
-        "btn = widgets.Button(description=\"Predict brain map\", button_style=\"primary\", icon=\"play\")\n",
-        "out = widgets.Output()\n",
-        "\n",
-        "PRESETS = {\n",
-        "    \"Sunny park\": \"a dog running through a sunny park\",\n",
-        "    \"Scary scene\": \"a terrifying monster jumps out in a dark alley\",\n",
-        "    \"Romantic\": \"two people sharing a quiet candlelit dinner\",\n",
-        "    \"Math puzzle\": \"solving a difficult calculus proof step by step\",\n",
-        "}\n",
-        "preset_btns = [widgets.Button(description=k, layout=widgets.Layout(width=\"140px\")) for k in PRESETS]\n",
-        "\n",
-        "def _on_preset(btn):\n",
-        "    text_in.value = PRESETS[btn.description]\n",
-        "\n",
-        "for b in preset_btns:\n",
-        "    b.on_click(_on_preset)\n",
-        "\n",
-        "def _on_predict(_):\n",
-        "    with out:\n",
-        "        clear_output(wait=True)\n",
-        "        text = text_in.value.strip()\n",
-        "        if not text:\n",
-        "            print(\"Enter some text first.\")\n",
-        "            return\n",
-        "        print(f\"Predicting for: {text!r} …\")\n",
-        "        rois = predict_rois(text)\n",
-        "        plot_results(text, rois)\n",
-        "\n",
-        "btn.on_click(_on_predict)\n",
-        "\n",
-        "display(widgets.VBox([\n",
-        "    text_in,\n",
-        "    widgets.HBox([btn] + preset_btns),\n",
-        "    out,\n",
-        "]))"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {},
-      "source": [
-        "## What the output means\n",
-        "\n",
-        "| Output | Description |\n",
-        "|--------|-------------|\n",
-        "| **Bar chart** | Top 16 of 56 ROI activation scores (positive = predicted up-regulation) |\n",
-        "| **Brain surface** | ROI values painted onto **fsaverage5** via HCP-MMP1 parcels (MNE + Nilearn) |\n",
-        "| **Skipped ROIs** | Subcortical regions (amygdala, striatum, thalamus) are not on the cortical surface |\n",
-        "\n",
-        "**Links:** [Model on HF](https://huggingface.co/ffh92r32rm0/Amphora_NeuroText) · [Code](https://github.com/hamcoderfran/neuroevolution)"
-      ]
-    }
-  ],
-  "metadata": {
-    "colab": {
-      "provenance": [],
-      "gpuType": "T4"
-    },
-    "kernelspec": {
-      "display_name": "Python 3",
-      "name": "python3"
-    },
-    "language_info": {
-      "name": "python"
-    },
-    "accelerator": "GPU"
-  },
-  "nbformat": 4,
-  "nbformat_minor": 5
-}