Probabilistic Switching Task (freely moving)¶

Protocol name: probabilistic_switching Setup: Freely moving, 3-port Bpod panel (left / centre / right) Video: Optional (RPi camera ensemble via config_file_camera)

Behavioural paradigm¶

A two-armed bandit task in which the mouse must track which of two side ports (left or right) is currently "rich": i.e. delivers water on the majority of choices. Reward probabilities switch covertly in blocks; the mouse must detect the switch and update its choice behaviour accordingly.

This class of task is widely used to study flexible decision making, reversal learning, and the contributions of prefrontal cortex and basal ganglia to value-based choice under uncertainty. The block structure with probabilistic (rather than deterministic) feedback prevents simple win-stay/lose-shift strategies and requires integration over multiple trials.

Block structure¶

Reward probabilities are assigned from the probabilities list in task.yaml. Each entry is a pair [p_left, p_right] (percentage, 0-100). Example defaults:

| Block type | p(reward | left) | p(reward | right) | |---|---|---| | Contrast high-left | 90% | 10% | | Contrast high-right | 10% | 90% | | Neutral | 50% | 50% |

Blocks are drawn without replacement (excluding the current block type unless block_switch_to_different_block_type = false). The first block is biased toward an easy contrast block (first_block_easy = true) to motivate the subject at session start.

Block switching criterion¶

Blocks switch when both conditions are met:

Minimum length: at least min_block_length trials have elapsed in the current block
Criterion: the exponential moving average of choices (EMA, time-constant criterion_tau trials) exceeds the criterion_contrast_blocks threshold toward the rich side

After the criterion is first met, the block continues for at least min_trials_post_criterion additional trials before the switch is enacted, adding stochasticity to the switch point and preventing the animal from exploiting a perfectly predictable reversal timing.

Neutral blocks switch after a random number of trials drawn from a geometric distribution (mean mean_neutral_block_length, minimum min_block_length).

Trial structure¶

centre-port init  →  side choice (left / right)
                            ↓ rewarded side
                        reward  →  ITI
                            ↓ non-rewarded side
                        no reward  →  ITI
                            ↓ timeout
                        timeout  →  ITI

The mouse initiates each trial by poking the centre port
After a short delay (delay_until_center_init) the side ports become active
Choice window: delay_until_side_timeout seconds; timeout → no reward
ITI: drawn from a uniform distribution (configurable)

Performance tracking¶

An exponential moving average (EMA) of choices tracks the animal's bias toward left (-1) or right (+1) with time-constant criterion_tau. The EMA is used for: - Block-switch criterion (see above) - Online plot display - Forced-choice anti-bias correction

Forced-choice anti-bias¶

When the animal has made forced_choice_threshold consecutive choices to the same side, the next trial is a forced-choice trial: only one side port is active, directing the animal to the opposite side. This prevents strong side biases from dominating performance metrics.

Key parameters¶

Parameter	Default	Description
`probabilities`	`[[10,50],[50,10],[10,90],[90,10],[50,50],...]`	Block reward probability pairs (p_left, p_right in %)
`min_block_length`	20	Minimum trials before block switch
`min_trials_post_criterion`	5	Minimum extra trials after criterion before switch
`mean_neutral_block_length`	25	Mean length of neutral blocks
`criterion_contrast_blocks`	0.75	EMA threshold to trigger contrast block switch
`criterion_neutral_blocks`	0.25	EMA threshold (not used for switch; informational)
`criterion_tau`	5	EMA time-constant (trials)
`first_block_easy`	true	Start session with a high-contrast block
`block_switch_to_different_block_type`	true	Exclude current block type when drawing next
`reset_bias_on_block_switch`	false	Reset EMA to 0 at each block switch
`forced_choice_threshold`	20	Consecutive same-side choices before forced trial
`reward_amount_ul`	2.0 µL	Water per rewarded choice
`n_max_trials`	1500	Hard trial limit
`delay_until_center_init`	[0.15, 0.2, 0.005]	Centre-port init delay (min, max, step; s)
`delay_until_side_timeout`	10 s	Side-port choice window

Modes¶

Mode	Description
(none defined)	All sessions use the same default block structure

Site-specific probability sets (e.g. omitting neutral blocks) can be set via a config-dir overlay or -ts flags.

Online plot¶

Moving average of choices (left/right bias)
Block probability trace (left / right)
Per-trial outcome scatter (rewarded / unrewarded / stop-signal)
Session trial count

Running¶

msw run -t probabilistic_switching -s mouse001 --setup setup-1
# With camera video:
msw run -t probabilistic_switching -s mouse001 --setup setup-1 \
    -ts config_file_camera=/data/msw_configs/cameras/setup-1.yaml