Merge branch 'main' into dtw_clean

Author: edyoshikun

.readthedocs.yaml

@@ -25,4 +25,4 @@ python:
 
 # Build documentation in the "docs/" directory with Sphinx
 sphinx:
-   configuration: docs/conf.py
+   configuration: docs/conf.py

applications/benchmarking/DynaCLR/ImageNet/imagenet_embeddings.py

@@ -50,7 +50,7 @@ def __init__(
             self.model = timm.create_model(model_name, pretrained=True)
             self.model.eval()
         except ImportError:
-            raise ImportError("Please install the timm library: " "pip install timm")
+            raise ImportError("Please install the timm library: pip install timm")
 
     def _reduce_5d_input(self, x: torch.Tensor) -> torch.Tensor:
         """Reduce 5D input (B, C, D, H, W) to 4D (B, C, H, W) using specified methods.
@@ -303,7 +303,7 @@ def main(config, model):
             phate_kwargs = cfg["embedding"]["phate_kwargs"]
 
         if "umap_kwargs" in cfg["embedding"]:
-            umap_kwargs = cfg["embedding"]["umap_kwargs"]
+            cfg["embedding"]["umap_kwargs"]
 
         if "pca_kwargs" in cfg["embedding"]:
             pca_kwargs = cfg["embedding"]["pca_kwargs"]

applications/benchmarking/DynaCLR/OpenPhenom/config_template.yml

@@ -13,11 +13,11 @@ model:
   # Default is "middle_slice" if not specified
   channel_reduction_methods:
     "Phase3D": "middle_slice"  # For phase contrast, middle slice often works well
-    "raw GFP EX488 EM525-45": "max"    
+    "raw GFP EX488 EM525-45": "max"
 
 # Data module configuration
 datamodule:
-  source_channel: 
+  source_channel:
     - Phase3D
     - "raw GFP EX488 EM525-45"
   z_range: [25, 40]
@@ -63,4 +63,4 @@ embedding:
 execution:
   overwrite: false
   save_config: true
-  show_config: true
+  show_config: true

applications/benchmarking/DynaCLR/OpenPhenom/openphenom_embeddings.py

@@ -57,8 +57,7 @@ def __init__(
             self.model.eval()
         except ImportError:
             raise ImportError(
-                "Please install the OpenPhenom dependencies: "
-                "pip install transformers"
+                "Please install the OpenPhenom dependencies: pip install transformers"
             )
 
     def on_predict_start(self):

applications/contrastive_phenotyping/evaluation/ALFI_MSD_v2.py

@@ -1,7 +1,9 @@
 # %%
 from pathlib import Path
+
 import matplotlib.pyplot as plt
 import numpy as np
+
 from viscy.representation.embedding_writer import read_embedding_dataset
 from viscy.representation.evaluation.distance import (
     compute_displacement,

applications/contrastive_phenotyping/evaluation/PC_vs_computed_features.py

@@ -1,13 +1,14 @@
-""" Script to compute the correlation between PCA and UMAP features and computed features
+"""Script to compute the correlation between PCA and UMAP features and computed features
 * finds the computed features best representing the PCA and UMAP components
 * outputs a heatmap of the correlation between PCA and UMAP features and computed features
 """
 
 # %%
 from pathlib import Path
+
 import matplotlib.pyplot as plt
 import seaborn as sns
-from compute_pca_features import compute_features, compute_correlation_and_save_png
+from compute_pca_features import compute_correlation_and_save_png, compute_features
 
 # %% for sensor features
 
@@ -44,7 +45,7 @@
 # features_sensor = pd.read_csv("/hpc/projects/comp.micro/infected_cell_imaging/Single_cell_phenotyping/ContrastiveLearning/Figure_panels/cell_division/features_allset_sensor.csv")
 
 # take a subset without the 768 features
-feature_columns = [f"feature_{i+1}" for i in range(768)]
+feature_columns = [f"feature_{i + 1}" for i in range(768)]
 features_subset_sensor = features_sensor.drop(columns=feature_columns)
 correlation_sensor = compute_correlation_and_save_png(
     features_subset_sensor,

applications/contrastive_phenotyping/evaluation/compute_pca_features.py

@@ -107,7 +107,7 @@ def compute_features(
     # convert the xarray to dataframe structure and add columns for computed features
     embedding_df = embedding_dataset["sample"].to_dataframe().reset_index(drop=True)
     feature_columns = pd.DataFrame(
-        features_npy, columns=[f"feature_{i+1}" for i in range(768)]
+        features_npy, columns=[f"feature_{i + 1}" for i in range(768)]
     )
 
     embedding_df = pd.concat([embedding_df, feature_columns], axis=1)
@@ -167,7 +167,6 @@ def compute_features(
     unique_fov_names = sorted(list(set(fov_names_list)))
 
     for fov_name in unique_fov_names:
-
         unique_track_ids = embedding_df[embedding_df["fov_name"] == fov_name][
             "track_id"
         ].unique()
@@ -180,7 +179,6 @@ def compute_features(
                 (embedding_df["fov_name"] == fov_name)
                 & (embedding_df["track_id"] == track_id)
             ].empty:
-
                 prediction_dataset = dataset_of_tracks(
                     data_path,
                     tracks_path,
@@ -214,7 +212,6 @@ def compute_features(
                         & (embedding_df["track_id"] == track_id)
                     ]["t"]
                 ):
-
                     # Basic statistical features for both channels
                     phase_features = CellFeatures(phase[i], nucl_mask[i])
                     PF = phase_features.compute_all_features()

applications/contrastive_phenotyping/evaluation/cosine_dissimilarity_dataset.py

@@ -3,9 +3,14 @@
 from typing import Optional
 
 import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
 import seaborn as sns
-from sklearn.preprocessing import StandardScaler
 from numpy.typing import NDArray
+from scipy.optimize import minimize_scalar
+from scipy.stats import gaussian_kde
+from sklearn.preprocessing import StandardScaler
+from tqdm import tqdm
 
 from viscy.representation.embedding_writer import read_embedding_dataset
 from viscy.representation.evaluation.clustering import (
@@ -14,13 +19,6 @@
     rank_nearest_neighbors,
     select_block,
 )
-import numpy as np
-from tqdm import tqdm
-import pandas as pd
-
-from scipy.stats import gaussian_kde
-from scipy.optimize import minimize_scalar
-
 
 plt.style.use("../evaluation/figure.mplstyle")
 

applications/contrastive_phenotyping/evaluation/displacement.py

@@ -6,74 +6,112 @@
 
 from viscy.representation.embedding_writer import read_embedding_dataset
 from viscy.representation.evaluation.distance import (
-   calculate_normalized_euclidean_distance_cell,
-   compute_displacement_mean_std_full,
+    calculate_normalized_euclidean_distance_cell,
+    compute_displacement_mean_std_full,
 )
 
-# %% paths 
+# %% paths
 
 features_path_30_min = Path(
-   "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/time_interval/predict/feb_test_time_interval_1_epoch_178.zarr"
+    "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/time_interval/predict/feb_test_time_interval_1_epoch_178.zarr"
 )
 
-feature_path_no_track = Path("/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_random_sampling2/feb_fixed_test_predict.zarr")
+feature_path_no_track = Path(
+    "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_random_sampling2/feb_fixed_test_predict.zarr"
+)
 
-features_path_any_time = Path("/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_difcell_randomtime_sampling/Ver2_updateTracking_refineModel/predictions/Feb_2chan_128patch_32projDim/2chan_128patch_56ckpt_FebTest.zarr")
+features_path_any_time = Path(
+    "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_difcell_randomtime_sampling/Ver2_updateTracking_refineModel/predictions/Feb_2chan_128patch_32projDim/2chan_128patch_56ckpt_FebTest.zarr"
+)
 
 data_path = Path(
-   "/hpc/projects/intracellular_dashboard/viral-sensor/2024_02_04_A549_DENV_ZIKV_timelapse/8-train-test-split/registered_test.zarr"
+    "/hpc/projects/intracellular_dashboard/viral-sensor/2024_02_04_A549_DENV_ZIKV_timelapse/8-train-test-split/registered_test.zarr"
 )
 
 tracks_path = Path(
-   "/hpc/projects/intracellular_dashboard/viral-sensor/2024_02_04_A549_DENV_ZIKV_timelapse/8-train-test-split/track_test.zarr"
+    "/hpc/projects/intracellular_dashboard/viral-sensor/2024_02_04_A549_DENV_ZIKV_timelapse/8-train-test-split/track_test.zarr"
 )
 
 # %% Load embedding datasets for all three sampling
-fov_name = '/B/4/6'
+fov_name = "/B/4/6"
 track_id = 52
 
 embedding_dataset_30_min = read_embedding_dataset(features_path_30_min)
 embedding_dataset_no_track = read_embedding_dataset(feature_path_no_track)
 embedding_dataset_any_time = read_embedding_dataset(features_path_any_time)
 
-#%%
+# %%
 # Calculate displacement for each sampling
-time_points_30_min, cosine_similarities_30_min = calculate_normalized_euclidean_distance_cell(embedding_dataset_30_min, fov_name, track_id)
-time_points_no_track, cosine_similarities_no_track = calculate_normalized_euclidean_distance_cell(embedding_dataset_no_track, fov_name, track_id)
-time_points_any_time, cosine_similarities_any_time = calculate_normalized_euclidean_distance_cell(embedding_dataset_any_time, fov_name, track_id)
+time_points_30_min, cosine_similarities_30_min = (
+    calculate_normalized_euclidean_distance_cell(
+        embedding_dataset_30_min, fov_name, track_id
+    )
+)
+time_points_no_track, cosine_similarities_no_track = (
+    calculate_normalized_euclidean_distance_cell(
+        embedding_dataset_no_track, fov_name, track_id
+    )
+)
+time_points_any_time, cosine_similarities_any_time = (
+    calculate_normalized_euclidean_distance_cell(
+        embedding_dataset_any_time, fov_name, track_id
+    )
+)
 
 # %% Plot displacement over time for all three conditions
 
 plt.figure(figsize=(10, 6))
 
-plt.plot(time_points_no_track, cosine_similarities_no_track, marker='o', label='classical contrastive (no tracking)')
-plt.plot(time_points_any_time, cosine_similarities_any_time, marker='o', label='cell aware')
-plt.plot(time_points_30_min, cosine_similarities_30_min, marker='o', label='cell & time aware (interval 30 min)')
+plt.plot(
+    time_points_no_track,
+    cosine_similarities_no_track,
+    marker="o",
+    label="classical contrastive (no tracking)",
+)
+plt.plot(
+    time_points_any_time, cosine_similarities_any_time, marker="o", label="cell aware"
+)
+plt.plot(
+    time_points_30_min,
+    cosine_similarities_30_min,
+    marker="o",
+    label="cell & time aware (interval 30 min)",
+)
 
 plt.xlabel("Time Delay (t)", fontsize=10)
 plt.ylabel("Normalized Euclidean Distance with First Time Point", fontsize=10)
-plt.title("Normalized Euclidean Distance (Features) Over Time for Infected Cell", fontsize=12)
+plt.title(
+    "Normalized Euclidean Distance (Features) Over Time for Infected Cell", fontsize=12
+)
 
 plt.grid(True)
 plt.legend(fontsize=10)
 
-#plt.savefig('4_euc_dist_full.svg', format='svg')
+# plt.savefig('4_euc_dist_full.svg', format='svg')
 plt.show()
 
 
 # %% Paths to datasets
-features_path_30_min = Path("/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/time_interval/predict/feb_test_time_interval_1_epoch_178.zarr")
-feature_path_no_track = Path("/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_random_sampling2/feb_fixed_test_predict.zarr")
+features_path_30_min = Path(
+    "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/time_interval/predict/feb_test_time_interval_1_epoch_178.zarr"
+)
+feature_path_no_track = Path(
+    "/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/time_sampling_strategies/negpair_random_sampling2/feb_fixed_test_predict.zarr"
+)
 
 embedding_dataset_30_min = read_embedding_dataset(features_path_30_min)
 embedding_dataset_no_track = read_embedding_dataset(feature_path_no_track)
 
 
 # %%
-max_tau = 10  
+max_tau = 10
 
-mean_displacement_30_min_euc, std_displacement_30_min_euc = compute_displacement_mean_std_full(embedding_dataset_30_min, max_tau)
-mean_displacement_no_track_euc, std_displacement_no_track_euc = compute_displacement_mean_std_full(embedding_dataset_no_track, max_tau)
+mean_displacement_30_min_euc, std_displacement_30_min_euc = (
+    compute_displacement_mean_std_full(embedding_dataset_30_min, max_tau)
+)
+mean_displacement_no_track_euc, std_displacement_no_track_euc = (
+    compute_displacement_mean_std_full(embedding_dataset_no_track, max_tau)
+)
 
 # %% Plot 2: Cosine Displacements
 plt.figure(figsize=(10, 6))
@@ -83,24 +121,44 @@
 mean_values_30_min_euc = list(mean_displacement_30_min_euc.values())
 std_values_30_min_euc = list(std_displacement_30_min_euc.values())
 
-plt.plot(taus, mean_values_30_min_euc, marker='o', label='Cell & Time Aware (30 min interval)', color='green')
-plt.fill_between(taus, 
-                 np.array(mean_values_30_min_euc) - np.array(std_values_30_min_euc), 
-                 np.array(mean_values_30_min_euc) + np.array(std_values_30_min_euc),
-                 color='green', alpha=0.3, label='Std Dev (30 min interval)')
+plt.plot(
+    taus,
+    mean_values_30_min_euc,
+    marker="o",
+    label="Cell & Time Aware (30 min interval)",
+    color="green",
+)
+plt.fill_between(
+    taus,
+    np.array(mean_values_30_min_euc) - np.array(std_values_30_min_euc),
+    np.array(mean_values_30_min_euc) + np.array(std_values_30_min_euc),
+    color="green",
+    alpha=0.3,
+    label="Std Dev (30 min interval)",
+)
 
 mean_values_no_track_euc = list(mean_displacement_no_track_euc.values())
 std_values_no_track_euc = list(std_displacement_no_track_euc.values())
 
-plt.plot(taus, mean_values_no_track_euc, marker='o', label='Classical Contrastive (No Tracking)', color='blue')
-plt.fill_between(taus, 
-                 np.array(mean_values_no_track_euc) - np.array(std_values_no_track_euc), 
-                 np.array(mean_values_no_track_euc) + np.array(std_values_no_track_euc),
-                 color='blue', alpha=0.3, label='Std Dev (No Tracking)')
+plt.plot(
+    taus,
+    mean_values_no_track_euc,
+    marker="o",
+    label="Classical Contrastive (No Tracking)",
+    color="blue",
+)
+plt.fill_between(
+    taus,
+    np.array(mean_values_no_track_euc) - np.array(std_values_no_track_euc),
+    np.array(mean_values_no_track_euc) + np.array(std_values_no_track_euc),
+    color="blue",
+    alpha=0.3,
+    label="Std Dev (No Tracking)",
+)
 
-plt.xlabel('Time Shift (τ)')
-plt.ylabel('Euclidean Distance')
-plt.title('Embedding Displacement Over Time (Features)')
+plt.xlabel("Time Shift (τ)")
+plt.ylabel("Euclidean Distance")
+plt.title("Embedding Displacement Over Time (Features)")
 
 plt.grid(True)
 plt.legend()

applications/contrastive_phenotyping/evaluation/imagenet_pretrained_features.py

@@ -111,7 +111,7 @@
 
 x_train = data_train_val.drop(
     columns=[
-         "division",
+        "division",
         "fov_name",
         "t",
         "track_id",