Scripts used to reproduce figures in: https://www.biorxiv.org/content/10.1101/2022.05.19.492731v2

• DataAccess.py: specs to download data from Allen Brain Observatory; a full guide is available at https://allensdk.readthedocs.io/en/latest/_static/examples/nb/ecephys_data_access.html
• Main_clean.m: plot Fig.2BCD.

• Build a python environment:
  conda env export > environment.yml

1. Train the models:

• Non-predictive model
  python Main_clean.py --input data/SeqN1T100.pth.tar --ae 1 -n 1 --fixo 3 --fixi 3 --rnn_act relu --ac_output sigmoid --savename Elman_SGD/Sigmoid/SeqN1T100_relu_fixio3
• Predictive model
  python Main_clean.py --input data/SeqN1T100.pth.tar --ae 1 -n 1 --fixi 3 --fixo 3 --pred 1 --rnn_act relu --ac_output sigmoid --savename Elman_SGD/Sigmoid/SeqN1T100_pred_relu_fixio3

2. Convert python output to *.mat
   python py2mat_ext.py

3. Plot the figure:
   MATLAB Figure3.m

1. Train the models: ./Figure4_run.sh

2. Plot the figures:
   python Figure4.py

1. Traj_Generate.m: simulation code for straight line exploration (Main reference: https://www.pnas.org/doi/10.1073/pnas.2018422118)

2. Localization_clean.py: generate model input from simulated trajectories.

3. Train the model:

• Non-predictive model
  for i in {1..10} do python Main_s4.py --epochs 30000 --batch-size 5 --hidden-n 500 --net ElmanRNN --act sigmoid --rnn_act relu --gpu 1 --Hregularized 5.0 --clip 10 --ae 1 --input data/InputNs50_SeqN100_StraightTraj_Marcus_v2.pth.tar --savename Elman_SGD/GridInput/BatchTraining/PhysicalInput_v2/repeats/InputNs50_SeqN100_Marcus_HN500_H5.0_rep$i done

• Predictive model
  for i in {1..10} do python Main_s4.py --epochs 30000 --batch-size 5 --hidden-n 500 --net ElmanRNN_tp1 --pred 1 --act sigmoid --rnn_act relu --gpu 4 --Hregularized 5.0 --clip 10 --ae 1 --input data/InputNs50_SeqN100_StraightTraj_Marcus_v2.pth.tar --savename Elman_SGD/GridInput/BatchTraining/PhysicalInput_v2/repeats/InputNs50_SeqN100_Marcus_HN500_H5.0_tp1_rep$i done

4. Plot the figure: python Figure5.py

1. Generate MNIST inputs:
   python Figure6_InputPrep.py

2. Train models:
   python Main.py -n 68 --input MNIST_68PC_SeqN100_Ns5.pth.tar --lr 0.0002 --pred 1 --partial 0.2 --ac_output tanh --Hregularized 1 --epochs 10000 -p 100 --savename Elman_SGD/predloss/MNIST_68PC_SeqN100_Ns5_partial

3. Plot the figure:
   python Figure6.py

1. Train models:
   python Main_local.py --n 68 --hidden_n 100 --inputfile 'data/data_pca.pkl' --savename 'mnist_local' --learning_alg 'local' --epochs 500001 --lr 0.001 --t 10

2. Plot the figure:
   python Figure7.py