Merge pull request #126 from lucywowen/master

code cleanup

.gitignore

@@ -101,6 +101,10 @@ ENV/
 # Rope project settings
 .ropeproject
 
+# pytest
+.pytest_cache
+
+# idea files
 .idea/**/
 
 *Untitled*

README.md

@@ -77,7 +77,7 @@ If you are reporting a bug, please do your best to include the following:
 
 <h3>Contributing code</h3>
 
-The preferred way to contribute to supereeg is to fork the main repository on GitHub, then submit a pull request.
+The preferred way to contribute to timecorr is to fork the main repository on GitHub, then submit a pull request.
 
 + If your pull request addresses an issue, please use the title to describe the issue and mention the issue number in the pull request description to ensure a link is created to the original issue.
 

tests/test_helpers.py

@@ -4,7 +4,6 @@
 import os
 import pytest
 
-#using method from supereeg
 from timecorr.helpers import gaussian_weights, gaussian_params, wcorr, wisfc, mat2vec, vec2mat, isfc, mean_combine, \
     corrmean_combine, timepoint_decoder, laplace_params, laplace_weights
 

tests/test_timecorr.py

@@ -3,7 +3,9 @@
 import pandas as pd
 import hypertools as hyp
 
+import timecorr as tc
 from timecorr.timecorr import timecorr
+from timecorr.simulate import simulate_data
 from timecorr.helpers import isfc, gaussian_weights, gaussian_params
 
 #TODO: need *real* tests-- e.g. generate a small dataset and verify that we actually get the correct answers
@@ -16,6 +18,68 @@
 # if above is how to make a numpy list than TC isn't capible np.lists currently
 random_numbers= (2 ,3 ,5, 10, 12, 4, 6)
 
+sim_1 = simulate_data(S=1, T=30, K=50, set_random_seed=100)
+sim_3 = simulate_data(S=3, T=30, K=50, set_random_seed=100)
+
+width = 10
+laplace = {'name': 'Laplace', 'weights': tc.laplace_weights, 'params': {'scale': width}}
+
+
+def test_reduce_shape():
+    dyna_corrs_reduced_1 = timecorr(sim_1, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'])
+
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'])
+    assert np.shape(dyna_corrs_reduced_1) == np.shape(sim_1)
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+def test_nans():
+    sim_3[0][0] = np.nan
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'])
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+
+def test_include_timepoints_all():
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'],
+                                    include_timepoints='all')
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+
+def test_include_timepoints_pre():
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'],
+                                    include_timepoints='pre')
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+
+def test_include_timepoints_post():
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'],
+                                    include_timepoints='post')
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+def test_exclude_timepoints_pos():
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'],
+                                    exclude_timepoints=3)
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
+
+def test_exclude_timepoints_neg():
+    dyna_corrs_reduced_3 = timecorr(sim_3, rfun='PCA',
+                                    weights_function=laplace['weights'], weights_params=laplace['params'],
+                                    exclude_timepoints=-3)
+
+    assert np.shape(dyna_corrs_reduced_3) == np.shape(sim_3)
+
 
 def test_timecorr():
 

timecorr/helpers.py

@@ -43,8 +43,6 @@ def laplace_weights(T, params=laplace_params):
 
 
 def eye_weights(T, params=eye_params):
-    #if params is None:
-    #    params = eye_params
 
     return np.eye(T)
 
@@ -153,7 +151,7 @@ def wisfc(data, timepoint_weights, subject_weights=None):
     if type(data) != list:
         return wisfc([data], timepoint_weights, subject_weights=subject_weights)[0]
 
-    if subject_weights is None: #similarity-based weights
+    if subject_weights is None:
         K = data[0].shape[1]
         connectomes = np.zeros([len(data), int((K ** 2 - K) / 2)])
         for s in np.arange(len(data)):
@@ -205,7 +203,7 @@ def apply_by_row(corrs, f):
     if type(corrs) is list:
         return list(map(lambda x: apply_by_row(x, f), corrs))
 
-    corrs = vec2mat(corrs) #V by V by T
+    corrs = vec2mat(corrs)
     return np.stack(list(map(lambda x: f(np.squeeze(x)), np.split(corrs, corrs.shape[2], axis=2))), axis=0)
 
 def corrmean_combine(corrs):
@@ -313,8 +311,15 @@ def reduce(corrs, rfun=None):
 
     if rfun in graph_measures.keys():
         return apply_by_row(corrs, graph_measures[rfun])
-    else:  # use hypertools
-        return hyp.reduce(corrs, reduce=rfun, ndims=V)
+    else:
+        red_corrs = hyp.reduce(corrs, reduce=rfun, ndims=V)
+
+        D = np.shape(red_corrs)[-1]
+
+        if D < V :
+            red_corrs = np.hstack((red_corrs, np.zeros((D, V - D))))
+
+        return red_corrs
 
 
 def smooth(w, windowsize=10, kernel_fun=laplace_weights, kernel_params=laplace_params):
@@ -667,7 +672,6 @@ def pca_decoder(data, nfolds=2, dims=10, cfun=isfc, weights_fun=laplace_weights,
     group_assignments = get_xval_assignments(len(pca_data), nfolds)
     results_pd = pd.DataFrame()
 
-    corrs = []
     for i in range(0, nfolds):
         for d in range(1, dims + 1):
 
@@ -799,37 +803,6 @@ def decoder(corrs):
     return next_results_pd
 
 
-# def predict(x, n=1):
-#     '''
-#     Use a Kalman filter (with automatically inferred parameters) to estimate
-#     future states of a signal, n timepoints into the future.
-#
-#     x: timepoints by features signal (numpy array)
-#     n: number of timepoints into the future (must be an integer)
-#
-#     Returns a new numpy array with x.shape[0] + n rows and x.shape[1] columns,
-#     where the last n rows contain the predicted future states.  The other
-#     entries contain "smoothed" estimates of the observed signals.
-#     '''
-#
-#     if n == 0:
-#         return kf.em(x).smooth(x)[0]
-#
-#     x_masked = np.ma.MaskedArray(np.vstack((x, np.tile(np.nan, (1, x.shape[1])))))
-#     x_masked[-1, :] = np.ma.masked
-#
-#     kf = pykalman.KalmanFilter(initial_state_mean=np.mean(x, axis=0), n_dim_obs=x.shape[1], n_dim_state=x.shape[1])
-#     x_predicted = kf.em(x_masked, em_vars='all').smooth(x_masked)
-#
-#     if n == 1:
-#         return x_predicted[0] #x_predicted[1] contains timepoint-by-timepoint covariance estimates
-#     elif n > 1:
-#         next_x_predicted = predict(x_predicted[0], n-1)
-#         diff = next_x_predicted.shape[0] - x_predicted[0].shape[0]
-#         next_x_predicted[:-diff, :] = x_predicted[0]
-#         return next_x_predicted
-
-
 def weighted_mean(x, axis=None, weights=None, tol=1e-5):
     if axis is None:
         axis = len(x.shape) - 1
@@ -975,14 +948,12 @@ def get_xval_assignments(ndata, nfolds):
     group_assignments = np.zeros(ndata)
     groupsize = int(np.ceil(ndata / nfolds))
 
-    # group assignments
     for i in range(1, nfolds):
         inds = np.arange(i * groupsize, np.min([(i + 1) * groupsize, ndata]))
         group_assignments[inds] = i
     np.random.shuffle(group_assignments)
     return group_assignments
 
-# set some defaults for plots
 SMALL_SIZE = 18
 MEDIUM_SIZE = 21
 BIGGER_SIZE = 24