fixed sample_args -> sample_params

scripts/example-workflow.py

@@ -8,46 +8,46 @@
 
 
 ########
-## 1. choose the hyper-paramters
+# 1. choose the hyper-paramters
 ########
 
 # define the hyper-parameter ranges
 ARGS = {
     # network architecture
     "pooling_layer": {"sum"},
-    "nonlinearity": {"relu"}, # relu or exp
+    "nonlinearity": {"relu"},  # relu or exp
 
     # motif
     "motif_length": {9},
     "n_motifs": {3},        # number of motifs
-    "init_motifs": {('TATTTAT', 'TTAATGA', 'ACTAAT')}, # seed motifs
+    "init_motifs": {('TATTTAT', 'TTAATGA', 'ACTAAT')},  # seed motifs
 
     # positional bias
     "n_splines": {10},    # number of b-splines to use. None means don't use the positional bias
-    "spline_exp": {False}, 
+    "spline_exp": {False},
     "share_splines": {False},   # should the positional bias be shared among different motifs
-    
+
     # parameters used by concise.prepare_data
     "trim_seq_len": {500},  # how many bp of the sequence should we use, None means whole
     "seq_align": {"end"},   # how to align sequences
-    
+
     # fitting parameteres
     "batch_size": 32,
     "n_epochs": 50,
-    "step_size": [0.001, 0.015], # Learning rate - Important hyper-parameter
+    "step_size": [0.001, 0.015],  # Learning rate - Important hyper-parameter
     "step_epoch": 10,
     "step_decay": (0.9, 1),
-    
+
     # regularization
     "lamb": [1e-7, 1e-3],
     "motif_lamb": [1e-9, 1e-3],
     "spline_lamb": [1e-9, 1e-3],
     "spline_param_lamb": [1e-9, 1e-3],
-    
+
     # weights scale (alternative to defining multiple different step_size)
     "nonlinearity_scale_factor": [1e-3, 1e3],
     "init_motifs_scale": [1e-2, 1e2],
-    
+
     # initialization
     "init_bias": {None},   # None means use the offset method
     "init_sd_filter": [1e-3, 1e-1],
@@ -56,12 +56,11 @@
 }
 
 # randomly sample a point the hyper-parameter space
-param = concise.sample_args(ARGS)
-
+param = concise.sample_params(ARGS)
 
 
 ########
-## 2. Import the data
+# 2. Import the data
 ########
 
 # mRNA half-life data from S. Pombe (Eser, Wachutka et.al. 2016 MSB):
@@ -76,21 +75,20 @@
                                                 id_column="ID",  # ID column name
                                                 seq_align=param['seq_align'],
                                                 trim_seq_len=param['trim_seq_len']
-                                              )
-
+                                                )
 
 
 ########
-## 3. Train the CONCISE model
+# 3. Train the CONCISE model
 ########
 
 # initialize with parameters defined previously
-co = concise.Concise(**param) 
+co = concise.Concise(**param)
 
 # train the model
-co.train(X_feat[500:],    X_seq[500:],    y[500:],     # training data
+co.train(X_feat[500:], X_seq[500:], y[500:],     # training data
          X_feat[100:500], X_seq[100:500], y[100:500],  # validation data
-         n_cores = 3)
+         n_cores=3)
 
 # predict for the test-set
 co.predict(X_feat[:100], X_seq[:100])
@@ -99,37 +97,36 @@
 co.get_weights()
 
 # save to file
-filepath  = "Concise_{0}.json".format(uuid.uuid4())
+filepath = "Concise_{0}.json".format(uuid.uuid4())
 co.save(filepath)
 
 # load from a file
 co2 = concise.Concise.load(filepath)
 
 
-
 ########
-## 4. Train the CONCISE model in cross-validation
+# 4. Train the CONCISE model in cross-validation
 ########
 
 # run the model in cross-validation and save the results
 
-# initialize the object 
+# initialize the object
 co = concise.Concise(**param)   # use the same parameters as before
-cocv = concise.ConciseCV(concise_model = co)
+cocv = concise.ConciseCV(concise_model=co)
 
 # train Concise in 5-fold cross-validation
 cocv.train(X_feat, X_seq, y, id_vec,
            n_folds=5, n_cores=3, train_global_model=True)
 
 
 # save to file
-filepath_cv  = "ConciseCV_{0}.json".format(uuid.uuid4())
+filepath_cv = "ConciseCV_{0}.json".format(uuid.uuid4())
 cocv.save(filepath_cv)
 
 # load from file
 cocv2 = ConciseCV.load(filepath_cv)
 
-## compute the mean-squared error on out-of-fold samples
+# compute the mean-squared error on out-of-fold samples
 y_cv_pred = cocv2.get_CV_prediction()
 np.mean((y - y_cv_pred)**2)