readme predict

README.md

@@ -94,7 +94,7 @@ The output of the GED computation on a set of graphs is a pickled list object co
 You can convert this output to a distance matrix and a list indicating the graphs each entry in the distance matrix correspond to.
 
 ```python
->>> from RNAmigos.post_ged import data_prepare
+>>> from RNAmigos.post_ged import prepare_data 
 >>> geds = '../data/geds_delta.pickle'
 >>> fps = '../data/all_rna_ligands_fingerprints.pickle'
 >>> DM, L, graphlist = prepare_data(geds, fps)
@@ -149,10 +149,23 @@ array([ 8., 10., 10.,  8., 22., 10.,  6.,  8.,  8., 18., 20.,  8., 10.,
 
 ### Embedding a single graph
 ```python
->>> G = nx.read_gpickle('/data/1jau.nxpickle')
->>> prototoypes = pickle.load('/data/sample_prototoypes.pickle')
->>> graph_embed(G, prototypes)
-array([1., 0., 2., 4., 1., 4., 5.])
+>>> G = nx.read_gpickle('data/1jau.nxpickle')
+>>> prototoypes = pickle.load('data/sample_prototoypes.pickle')
+>>> x = graph_embed(G, prototypes)
+>>> x
+array([ 0, 21, 17, 19, 17, 17, 13, 12, 11, 17, 14, 18, 17, 11, 17, 15,  7,
+       11, 19,  9, 19,  4, 14, 15, 23, 18, 18,  7, 19, 18, 16, 24, 23, 16,
+       16, 16, 24, 19, 13, 15, 14, 12, 13, 21, 17, 16, 17, 17,  6, 12, 13,
+       14,  2, 18, 19, 18, 16, 20, 19, 19, 23,  2, 19, 19,  4,  3,  4, 17,
+       18, 18,  9, 15, 17, 19, 21, 22, 18,  1, 21, 21, 16, 15, 18, 10,  2,
+       11,  0, 14, 19, 10, 15, 18, 14, 13, 11, 23,  1,  0,  3,  6,  2, 14,
+       21, 19, 22, 19, 20, 16, 17,  1, 20, 18, 11,  0, 10,  1,  3, 21, 21,
+       10, 27, 16, 14,  4,  5,  0, 14, 18, 12, 15, 18,  1,  1, 18,  6, 20,
+        0, 19, 17, 23, 14, 19, 16, 17, 17, 21, 21, 29, 16, 18, 18, 20, 19,
+       16, 18,  1,  1,  1, 18,  3,  0, 15, 21, 18, 15, 17, 12, 18,  6,  1,
+        8,  3,  6, 21, 19,  1, 15, 22, 14, 20, 20, 20, 16, 20, 16,  9,  5,
+        5, 13,  6])
+
 
 ```
 
@@ -163,7 +176,24 @@ User-friendly API coming soon.
 Once all graphs are embedded we have the standard machine learning input matrix $X$ with $n$ examples as rows and $r$ features as distances to each prototype. 
 Any type of classification can now be performed using a label (output) vector for single-class classification or matrix for multi-output classification.
 
+Alternatively, we can classify graphs using k-nearest neighbours and skip the embedding procedure.
 
+You can load a pre-trained model in `/models/` and embed new graphs to make predictions:
 
 
-Alternatively, we can classify graphs using k-nearest neighbours and skip the embedding procedure.
+```python
+>>> mlp = pickle.load(open('models/mlp_pre_train_spanning_190.pickle', 'rb'))
+>>> x = np.reshape((1, 190))
+>>> mlp.predict(x)
+array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+        0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0,
+        0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0,
+        0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 1,
+        0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1,
+        1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1, 0]])
+```
+
+The result is a 166 bit MACSS molecular fingerprint for the given graph embedding.
+

RNAmigos/dissimilarity_embed.py

@@ -200,7 +200,7 @@ def graph_embed(G, prototypes):
 
     `array`: numpy array representing embedding vector.
     """
-    embedding = np.zeros(len(prototypes))
+    embedding = []
     for p in prototypes:
         ops,_,_ = ged((G,p), source_only=True)
         embedding.append(ops.cost)