fix path concatenations (#26)

* fix path concatenations

* refactor

* update notebook

* update notebook

* update notebook

* update notebook

efaar_benchmarking/data_loading.py

@@ -35,7 +35,7 @@ def load_cpg16_crispr(data_path: str = "data/") -> tuple[pd.DataFrame, pd.DataFr
     well_file_path = os.path.join(data_path, well_file_name)
     crispr_file_path = os.path.join(data_path, crispr_file_name)
     if not (os.path.exists(plate_file_path) and os.path.exists(well_file_path) and os.path.exists(crispr_file_path)):
-        path_to_zip_file = data_path + "tmp.zip"
+        path_to_zip_file = os.path.join(data_path, "tmp.zip")
         wget.download(metadata_source_path, path_to_zip_file)
         with zipfile.ZipFile(path_to_zip_file, "r") as zip_ref:
             for name in zip_ref.namelist():
@@ -118,9 +118,10 @@ def load_replogle(gene_type: str, data_type: str, data_path: str = "data/") -> s
     else:
         raise ValueError("gene_type must be either essential or genome_wide")
 
-    if not os.path.exists(data_path + filename):
-        wget.download(src, data_path + filename)
+    fn = os.path.join(data_path, filename)
+    if not os.path.exists(fn):
+        wget.download(src, fn)
 
-    adata = sc.read_h5ad(data_path + filename)
+    adata = sc.read_h5ad(fn)
     adata = adata[:, np.all(~np.isnan(adata.X) & ~np.isinf(adata.X), axis=0)]
     return adata

efaar_benchmarking/efaar.py

@@ -25,6 +25,7 @@ def embed_by_scvi_anndata(adata, batch_col=cst.REPLOGLE_BATCH_COL, n_latent=128,
     Returns:
         numpy.ndarray: Embedding of the input data using scVI.
     """
+    adata = adata.copy()
     SCVI.setup_anndata(adata, batch_key=batch_col)
     vae = SCVI(adata, n_hidden=n_hidden, n_latent=n_latent)
     vae.train(use_gpu=True)
@@ -47,6 +48,30 @@ def embed_by_pca_anndata(adata, n_latent=100) -> np.ndarray:
     return adata.obsm["X_pca"]
 
 
+def centerscale(features: np.ndarray, metadata: pd.DataFrame = None, plate_col: Optional[str] = None) -> np.ndarray:
+    """
+    Center and scale the input features.
+
+    Args:
+        features (np.ndarray): Input features to be centered and scaled.
+        metadata (pd.DataFrame): Metadata information for the input features.
+        plate_col (str): Name of the column in metadata that contains plate information.
+
+    Returns:
+        np.ndarray: Centered and scaled features.
+    """
+    if plate_col is None:
+        features = StandardScaler().fit_transform(features)
+    else:
+        if metadata is None:
+            raise ValueError("metadata must be provided if plate_col is not None")
+        unq_plates = metadata[plate_col].unique()
+        for plate in unq_plates:
+            ind = metadata[plate_col] == plate
+            features[ind, :] = StandardScaler().fit_transform(features[ind, :])
+    return features
+
+
 def embed_align_by_pca(
     features: np.ndarray,
     metadata: pd.DataFrame = None,
@@ -58,6 +83,7 @@ def embed_align_by_pca(
     Note that we explicitly center & scale the data by plate before and after calling `PCA`.
     Centering and scaling is done by plate if `plate_col` is not None, and on the whole data otherwise.
     Note that `PCA` transformer also does mean-centering on the whole data prior to the PCA operation.
+
     Args:
         features (np.ndarray): Features to transform
         metadata (pd.DataFrame): Metadata. Defaults to None.
@@ -69,18 +95,6 @@ def embed_align_by_pca(
         np.ndarray: Transformed data using PCA.
     """
 
-    def centerscale(features, metadata, plate_col):
-        if plate_col is None:
-            features = StandardScaler().fit_transform(features)
-        else:
-            if metadata is None:
-                raise ValueError("metadata must be provided if plate_col is not None")
-            unq_plates = metadata[plate_col].unique()
-            for plate in unq_plates:
-                ind = metadata[plate_col] == plate
-                features[ind, :] = StandardScaler().fit_transform(features[ind, :])
-        return features
-
     features = centerscale(features, metadata, plate_col)
     features = PCA(variance_or_ncomp).fit_transform(features)
     features = centerscale(features, metadata, plate_col)

notebooks/replogle_map_building.ipynb

@@ -57,14 +57,19 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "3e9e8847",
-   "metadata": {},
+   "metadata": {
+    "scrolled": true
+   },
    "outputs": [],
    "source": [
     "adata = load_replogle(\"genome_wide\", \"normalized\")\n",
     "metadata = adata.obs\n",
+    "print(metadata.shape)\n",
     "embeddings = embed_by_pca_anndata(adata)\n",
+    "print(embeddings.shape)\n",
     "del adata\n",
     "embeddings = align_on_controls(embeddings, metadata)\n",
+    "print(embeddings.shape)\n",
     "map_data = aggregate(embeddings, metadata)\n",
     "del embeddings, metadata\n",
     "metrics = benchmark(map_data, recall_thr_pairs=recall_threshold_pairs)\n",
@@ -74,6 +79,7 @@
   },
   {
    "cell_type": "markdown",
+   "id": "2bd66d1d",
    "metadata": {},
    "source": [
     "## scVI Embeddings"
@@ -86,21 +92,26 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "adata = load_replogle(\"essential\", \"raw\")\n",
+    "adata = load_replogle(\"genome_wide\", \"raw\")\n",
     "metadata = adata.obs\n",
+    "print(metadata.shape)\n",
     "embeddings = embed_by_scvi_anndata(adata)\n",
+    "print(embeddings.shape)\n",
+    "del adata\n",
     "embeddings = align_on_controls(embeddings, metadata)\n",
     "map_data = aggregate(embeddings, metadata)\n",
+    "del embeddings, metadata\n",
     "metrics = benchmark(map_data, recall_thr_pairs=recall_threshold_pairs)\n",
-    "plot_recall(metrics)"
+    "plot_recall(metrics)\n",
+    "metrics.groupby('source')['recall_0.05_0.95'].mean()"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "eben",
    "language": "python",
-   "name": "python3"
+   "name": "eben"
   },
   "language_info": {
    "codemirror_mode": {
@@ -112,7 +123,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.5"
+   "version": "3.10.4"
   }
  },
  "nbformat": 4,