Merge pull request #6 from gyoge0/vmap_generate_trace

add vmapped version of generate_trace

blinx/trace_model.py

@@ -30,9 +30,13 @@ def log_p_parameters(parameters, locs, scales):
     if locs.sigma_ro is not None:
         log_p += jnp.log(norm.pdf(parameters.sigma_ro, locs.sigma_ro, scales.sigma_ro))
     if locs._p_on_logit is not None:
-        log_p += jnp.log(norm.pdf(parameters.p_on, locs._p_on_logit, scales._p_on_logit))
+        log_p += jnp.log(
+            norm.pdf(parameters.p_on, locs._p_on_logit, scales._p_on_logit)
+        )
     if locs._p_off_logit is not None:
-        log_p += jnp.log(norm.pdf(parameters.p_off, locs._p_off_logit, scales._p_off_logit))
+        log_p += jnp.log(
+            norm.pdf(parameters.p_off, locs._p_off_logit, scales._p_off_logit)
+        )
 
     return log_p
 
@@ -281,6 +285,51 @@ def sample_next_z(z, p_transition, key):
     return z
 
 
+def vmap_generate_trace(
+    num_traces, y, parameters, num_frames, hyper_parameters, seed=None
+):
+    """Create several simulated intensity traces.
+
+    Args:
+        num_traces (int):
+            - the number of traces to simulate
+
+        y (int):
+            - the total number of fluorescent emitters
+
+        parameters (:class:'Parameters'):
+            - the parameters of the fluoresent and trace model
+
+        num_frames (int):
+            - the number of observations to simulate
+
+        hyper_parameters (:class:`HyperParameters`):
+            - hypxer-parameters with `delta_t` set for the time between frames in the traces
+
+        seed (int, optional):
+            - random seed for the jax psudo rendom number generator
+
+    Returns:
+        trace (array):
+            - a num_traces x num_frames array containing traces with intensity values for each frame
+
+        states (array):
+            - array the same shape as trace, containing the number of 'on' emitters in each frame
+    """
+
+    if seed is None:
+        seed = time.time_ns()
+    key = random.PRNGKey(seed)
+    subkeys = random.split(key, num_traces)
+    seeds = subkeys[:, 0]
+    mapped = jax.vmap(
+        generate_trace,
+        in_axes=(None, None, None, None, 0),
+    )
+    trace, zs = mapped(y, parameters, num_frames, hyper_parameters, seeds)
+    return jnp.squeeze(trace), jnp.squeeze(zs)
+
+
 def create_transition_matrix(y, p_on, p_off):
     """Create a transition matrix for the number of active elements, given that
     elements can randomly turn on and off.