README for project file Schapiro_etal_hippocampus_model.proj used in:
Schapiro, Turk-Browne, Botvinick, & Norman. (2016). Complementary learning systems within the hippocampus: A neural network modeling approach to reconciling episodic memory with statistical learning. Phil. Trans. R. Soc. B. doi: 10.1098/rstb.2016.0049.

Schapiro_etal_hippocampus_model.proj was developed in version 7.0.1 (7740) of Emergent (http://grey.colorado.edu/emergent). Note that this is not the most recent version and can be downloaded from here: https://grey.colorado.edu/emergent_ftp/.

To run simulations in this project, download the above version of Emergent. Open the project file in Emergent, and click on the ControlPanel tab in the center pane.

There are three networks in this project file: SLnet_pairs, SLnet_associative_inference, and SLnet_community, corresponding to the three networks reported in Schapiro et al. (2016). Choose a ‘network’ in the ControlPanel, and then corresponding ‘train_patterns’ and ‘test_patterns’. 

You can specify the filename for the write-out of unit activities in ‘writeout_filename’. Unit activites evoked by the test items are recorded in separate files for each epoch and batch and for the initial and settled response.

Then specify the ‘num_trials_per_epoch’, ‘num_epochs_per_batch’, and ‘num_batches’. In the paper:
- Every simulation was run with 500 batches (re-initializations of the projections and weights).
- Pair structure had 80 trials per epoch, for 10 epochs. 
- Pair structure without transitions (the ‘episodic’ simulation) had 4 trials per epoch, for 10 epochs.
- Community structure had 60 trials per epoch, for 10 epochs.
- Associative inference had 60 trials per epoch, for 20 epochs. (There is also a variant in InputData with only 6 trials, which we used for the simulations of one exposure to each of the direct pairs.)

Checking ‘run_test_during_training’ means that the software will automatically run a test and save the unit activities before each epoch of training.

Checking ‘use_low_inhibition_at_test’ will raise the k value during test to allow more units to be active. See the program code in LeabraEpochTest to view and modify specific k values.

‘max_test_cycles’ controls the number of cycles allowed per test trial if you are running the test functions through the ControlPanel. (If using ‘run_test_during_training’, this value will be set automatically to 20 to test the initial response and 80 to test the settled response to each test item.)

The next six variables control the strength of the CA3->CA1 and ECin->CA1 projections for each of the three networks. Reducing the value of the CA3->CA1 projection is the simplest way to implement a TSP lesion (as this is the pathway that communicates the ouput of the TSP to the rest of the network), and reducing the value of the ECin->CA1 projection implements an MSP pathway lesion.

To train the network, Click ‘Batch_Init’ then ‘Batch_Run’ at the bottom of the ControlPanel. To test the network, click ’Test_Init’ then ‘Test_Run’.

Click the appropriate network tab in the right pane to view the activity of the network during training and test. To speed up simulations, turn off the display updates by clicking on the tab with the network name in the middle pane and unchecking ‘Display’.