KeyError in DataLoader when (tx, belongs, bd).edge_label is missing for some samples #94

Activity

1. mossishahi

   assigned

   andrewmoorman

   and

   EliHei2

   on Mar 24, 2025
2. andrewmoorman commented on Apr 1, 2025

   

   andrewmoorman

   on Apr 1, 2025

   Collaborator

   More actions

   Thanks for flagging this @mossishahi! I'll take a look.
3. BioinfoTongLI commented on Jun 9, 2025

   

   BioinfoTongLI

   on Jun 9, 2025

   Last edited by BioinfoTongLI

   More actions

   Hello there!
   I think I've got a similar issue with edge_label in the prediction step, which looks like the error message in this one #81 (comment)
   Any ideas what to try? @EliHei2
   both create dataset and train_model finished correctly

   python3 /workspace/segger_dev/src/segger/cli/create_dataset_fast.py \
       --base_dir output-XETG00155__0037475__AT9-BRA-4-FO-4-S51-ii__20241107__134704 \
       --data_dir AT9-BRA-4-FO-4-S51-ii_dataset \
       --sample_type xenium \
       --n_workers 30 \
       --tile_width 120 \
       --tile_height 120 \
       --scrnaseq_file GBM_LEAP_annotations_v2.09_04_24__AT9.h5ad \
       --celltype_column annotation_coarse             --sample_type xenium_v2             --k_bd 3             --dist_bd 15.0             --k_tx 3             --dist_tx 5.0             --neg_sampling_ratio 5.0             --frac 1.0             --val_prob 0
   .1             --test_prob 0.2
   

   python3 /workspace/segger_dev/src/segger/cli/train_model.py \
       --dataset_dir AT6-BRA-3-FO-3-S40-i_dataset \
       --models_dir AT6-BRA-3-FO-3-S40-i_trained_models \
       --sample_tag AT6-BRA-3-FO-3-S40-i \
       --num_workers 30 \
       --batch_size 4 \
       --max_epochs 200 \
       --devices 1 \
       --accelerator cuda \
       --sample_tag AT13-BRA-3-FO-4-S44-ii_scRNA_annotation_coarse             --init_emb 8             --hidden_channels 32             --out_channels 16             --heads 4             --num_mid_layers 3             --batch_size 2             --num_worke
   rs 2             --accelerator cuda             --max_epochs 100             --devices 1             --strategy auto             --precision 32
   

   but not prediction step.

   python3 /workspace/segger_dev/src/segger/cli/predict_fast.py \
       --models_dir AT6-BRA-3-FO-2-S39-ii_trained_models \
       --segger_data_dir AT6-BRA-3-FO-2-S39-ii_dataset \
       --transcripts_file transcripts.parquet \
       --benchmarks_dir AT6-BRA-3-FO-2-S39-ii_benchmarks_dir \
       --num_workers 30 \
       --batch_size 1 \
       --use_cc false \
       --knn_method kd_tree \
           --batch_size 2             --num_workers 2             --model_version 0             --save_tag AT13-BRA-3-FO-4-S44-ii_scRNA_annotation_coarse             --min_transcripts 5             --cell_id_col segger_cell_id             --use_cc false
           --knn_method kd_tree             --file_format anndata             --k_bd 4             --dist_bd 15.0             --k_tx 5             --dist_tx 5.0
   

   cc'ing @krademaker
   And I am using this docker images

   danielunyi42/segger_dev                                                cuda121                   e14dbf486142
   

   👍React with 👍1Reacted by Koen Rademaker