TST: batched stimulus computation

ENH: basic functionality for solving issue #1

moten/tests/test_api.py

@@ -1,16 +1,15 @@
 import numpy as np
-
 import matplotlib
-matplotlib.interactive(False)
 
 from moten import (io,
                    api,
+                   core,
                    utils,
                    )
 
 if 1:
     video_file = 'http://anwarnunez.github.io/downloads/avsnr150s24fps_tiny.mp4'
-    nimages = 100
+    nimages = 200
     small_size = (72, 128) # downsampled size (vdim, hdim) 16:9 aspect ratio
     luminance_images = io.video2luminance(video_file, size=small_size, nimages=nimages)
     #luminance_images = io.video2grey(video_file, size=small_size, nimages=nimages)
@@ -27,10 +26,12 @@
                                       spatial_frequencies=(0,1,2,4))
 print(pyramid_view)
 
+
 # stimulus pyramid
 ##############################
 pyramid = api.StimulusMotionEnergy(luminance_images,
                                    stimulus_fps,
+                                   filter_temporal_width=10,
                                    spatial_frequencies=(0,1,2,4))
 # pyramid constructor is an attribute
 print(pyramid.view)
@@ -61,3 +62,36 @@
 hvresponses_sin, hvresponses_cos = pyramid.raw_projection(hvcentered_filters, dtype=DTYPE)
 responses_manualhv = utils.log_compress(utils.sqrt_sum_squares(hvresponses_sin, hvresponses_cos))
 assert np.allclose(hvcentered_responses, responses_manualhv)
+
+# project specific stimuli
+filter_stimulus_responses = pyramid.project_stimulus(luminance_images,
+                                                     dtype=DTYPE)
+assert np.allclose(filter_stimulus_responses, filter_responses)
+
+
+##############################
+# stimulus batches
+##############################
+
+# project subset of original stimuli
+first_frame, last_frame = 100, 110
+filter_stimulus_responses = pyramid.project_stimulus(
+    luminance_images[first_frame:last_frame], dtype=DTYPE)
+
+# these differ because of convolution edge effects
+assert not np.allclose(filter_stimulus_responses,
+                       filter_responses[first_frame:last_frame])
+
+# we have to include a window in order to avoid edge effects.
+# This window is determined by the FOV of the motion-energy filter.
+# In this case, the filter is the same width as the stimulus frame
+# rate.
+filter_width = pyramid.view.definition.filter_temporal_width
+window = int(filter_width/2)
+windowed_filter_stimulus_responses = pyramid.project_stimulus(
+    luminance_images[first_frame - window:last_frame + window],
+    dtype=DTYPE)
+
+# Now they are exactly the same
+assert np.allclose(windowed_filter_stimulus_responses[window:-window],
+                   filter_responses[first_frame:last_frame])