Refactor notebook to handle the case of more than

2 classes. Include Github Action to get the data
and generate the embeddings with the Hyena Helical
model.

.github/workflows/get_embeddings.yml

@@ -0,0 +1,39 @@
+name: CI Pipeline
+
+on:
+  workflow_dispatch:
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v2
+
+      - name: setup python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.11.8
+          cache: 'pip' # caching pip dependencies
+
+      - name: Install dependencies
+        run: |
+            pip install .
+            
+      # First download before tests as they make use of the downloaded files 
+      - name: Download all files
+        run: |
+          python ci/download_all.py
+
+      - name: Execute script to get embeddings
+        run: |
+          python ci/get_all_data_embeddings.py
+      
+      - name: Upload numpy embeddings data
+        uses: actions/upload-artifact@v2
+        with:
+          name: upload-data
+          path: data/
+
+

ci/get_all_data_embeddings.py

@@ -0,0 +1,58 @@
+import numpy as np
+from datasets import DatasetDict
+from tqdm import tqdm
+from helical.models.hyena_dna.model import HyenaDNA
+from helical.models.hyena_dna.hyena_dna_config import HyenaDNAConfig    
+from datasets import get_dataset_config_names
+from datasets import load_dataset
+import os
+
+configurer = HyenaDNAConfig(model_name="hyenadna-tiny-1k-seqlen-d256")
+hyena_model = HyenaDNA(configurer=configurer)
+
+
+def get_model_inputs(dataset: DatasetDict, ratio: float = 1.0):
+
+    x = np.empty((0, configurer.config['d_model'])) 
+    labels = np.empty((0,), dtype=int)
+
+    # disable logging to avoid cluttering the output
+    import logging
+    logging.disable(logging.CRITICAL)
+
+    # use tqdm for a progress bar
+    length = int(len(dataset)*ratio)
+    for i in tqdm(range(length)):
+        sequence = dataset["sequence"][i]
+
+        tokenized_sequence = hyena_model.process_data(sequence)
+        embeddings = hyena_model.get_embeddings(tokenized_sequence)
+
+        numpy_array = embeddings[0].detach().numpy()
+        mean_array = numpy_array.mean(axis=0)
+        x = np.append(x, [mean_array], axis=0)
+
+    # normalize the data
+    x = (x - np.mean(x, axis=0)) / np.std(x, axis=0)
+    labels = np.array(dataset["label"][:length])
+    return x, labels
+
+labels = get_dataset_config_names("InstaDeepAI/nucleotide_transformer_downstream_tasks")
+
+for label in labels:
+    dataset = load_dataset("InstaDeepAI/nucleotide_transformer_downstream_tasks", label)
+
+    x, y = get_model_inputs(dataset["train"], 0.001)
+
+    if not os.path.exists("data"):
+        os.makedirs("data")
+
+    np.save(f"data/x_{label}_norm_256", x)
+    np.save(f"data/y_{label}_norm_256", y)
+
+    X_unseen, y_unseen = get_model_inputs(dataset["test"], 0.001)
+    np.save(f"data/x_unseen_{label}_norm_256", X_unseen)
+    np.save(f"data/y_unseen_{label}_norm_256", y_unseen)
+    break
+
+