Fix LSTM LRP bug

bokeh_vis/main.py

@@ -25,8 +25,9 @@
 
 
 def button_callback():
-    text_src = re.sub('/home/icha/','/home/yannis/Desktop/tRustNN/',rawInput_selections.value)
-    text_banner.text = open(text_src,"r").read()
+
+    text_review,words,word_embeddings = get_rawText_data(rawInput_selections.value,keys_raw,data_raw,testX,embed_mat)
+    text_banner.text = text_review
 
 def get_wc_colourGroups(rawInput_source):
 
@@ -61,13 +62,15 @@ def get_clustering_selections(algorithms_neurons):
 
     return (algorithm_select_neuron,cluster_slider)
 
+
 def get_rawInput_selections(keys_raw):
 
     review = [str(r) for r in list(keys_raw)]
     select_rawInput = Select(title="Input review", value=review[0], options=review)
 
     return select_rawInput
 
+
 def get_projection_selections(algorithms):
 
     algorithm_select = Select(value="PCA",title="Select projection algorithm:",width=250, options=algorithms)
@@ -77,12 +80,13 @@ def get_projection_selections(algorithms):
 
 def get_rawText_data(rawInput_selections,keys_raw,data_raw,feed,embed_mat):
 
-
-    text_review = str(data_raw[int(rawInput_selections)])
-    words = str(text_review).split()
+    text_review = np.array_str(data_raw[int(rawInput_selections)])
+    txt_rev = text_review.replace('<UNK>','UNK')
+
+    words = text_review.split()
     word_embeddings = [embed_mat[i,:] for i in  list(feed[int(rawInput_selections),:].astype(int))]
 
-    return text_review,words,word_embeddings
+    return txt_rev,words,word_embeddings
 
 
 """
@@ -148,7 +152,7 @@ def update_source(attrname, old, new):
         if algorithm_cl_neurons=="KMeans - selected gate":
             text_set.text = "KMeans: Clusters neurons based on their gate values after training."
         elif algorithm_cl_neurons=="DBSCAN - selected review":
-            text_set.text = "DBSCAN - selected review: Clusters neurons based on how related their most activating words are. List of activating words generated from seleceted review."
+            text_set.text = "DBSCAN - selected review: Clusters neurons based on how related their most activating words are. List of activating words generated from selected review."
         neuronData = similarityMatrix_PerReview
         cluster_labels, colors, _ = clustering.apply_cluster(x,algorithm_cl_neurons,n_clusters,review=int(rawInput_selections.value),neuronData=neuronData,mode="nn")
 
@@ -166,7 +170,7 @@ def update_source(attrname, old, new):
     elif gate_value=="output_gate":
         wc_filename,wc_img,wc_words = get_wcloud(LRP,int(rawInput_selections.value),load_dir,color_dict=color_dict,gate="out")
 
-    words_to_be_highlighted = [i for i in wc_words and totalLRP[int(rawInput_selections.value)]['words']]
+    words_to_be_highlighted = list(set(wc_words).intersection(totalLRP[int(rawInput_selections.value)]['words']))
     lrp_source.data['lrp'] = scaler.fit_transform(np.array(totalLRP[int(rawInput_selections.value)]['lrp'].tolist()).reshape(-1,1))
     tap_source.data['wc_words'] = words_to_be_highlighted
     wc_plot.add_glyph(img_source, ImageURL(url=dict(value=load_dir+wc_filename), x=0, y=0, anchor="bottom_left"))
@@ -196,10 +200,7 @@ def update_source(attrname, old, new):
 with open(load_dir+"exploratoryDataFull.pickle", 'rb') as f:
     (testX,embed_mat,excitingWords_fullSet,similarityMatrix_AllReviews,similarityMatrix_PerReview,neuron_types,totalLRP,LRP) = pickle.load(f)
 
-
-#neuronExcitingWords_AllReviews = list((excitingWords_fullSet.values()))
 _,lstm_hidden = data_format.get_data(load_dir+"test_model_internals_lstm_hidden.pickle")
-#_,learned_embeddings = data_format.get_data(load_dir+"test_model_internals_ebd.pickle")
 
 #Get preset buttons' selections
 
@@ -228,7 +229,7 @@ def update_source(attrname, old, new):
 project_plot.add_tools(taptool)
 
 #Input text
-text_review,words,word_embeddings = get_rawText_data(rawInput_selections.value,keys_raw,data_raw,testX,embed_mat) 
+text_review,words,word_embeddings = get_rawText_data(rawInput_selections.value,keys_raw,data_raw,testX,embed_mat)
 w2v_labels, w2v_colors, _ = clustering.apply_cluster(np.array(word_embeddings),algorithm="KMeans - selected gate",n_clusters=int(clustering_selections[1].value),mode="wc")
 rawInput_source = ColumnDataSource(dict(z=w2v_colors,w=words))
 
@@ -248,6 +249,7 @@ def update_source(attrname, old, new):
 tap_source = ColumnDataSource(dict(wc_words=words_to_be_highlighted))
 scaler = MinMaxScaler(copy=True, feature_range=(-1, 1))
 lrp_source = ColumnDataSource(dict(lrp=scaler.fit_transform(np.array(totalLRP[int(rawInput_selections.value)]['lrp'].tolist()).reshape(-1,1))))
+
 #totalLRP : how relevant is each LSTM neuron
 
 
@@ -286,7 +288,7 @@ def update_source(attrname, old, new):
 xdr = Range1d(start=0, end=600)
 ydr = Range1d(start=0, end=600)
 wc_plot = Plot(title=None, x_range=xdr, y_range=ydr, plot_width=500, plot_height=550, min_border=0)
-image = ImageURL(url=dict(value=load_dir+wc_filename), x=0, y=0, anchor="bottom_left", retry_attempts=5, retry_timeout=1500)
+image = ImageURL(url=dict(value=load_dir+wc_filename), x=0, y=0, anchor="bottom_left", retry_attempts=3, retry_timeout=1000)
 wc_plot.add_glyph(img_source, image)
 
 
@@ -297,10 +299,10 @@ def update_source(attrname, old, new):
 lrp_timedata = get_lrp_timedata(LRP)
 time = [i for i in range(len(lrp_timedata))]
 lrptime_source = ColumnDataSource(dict(lrptime = lrp_timedata,time=time))
-lrp_plot = figure(title="Total normalized LRP per timestep",plot_width=300, plot_height=50)
+lrp_plot = figure(title="Network focus per timestep",plot_width=300, plot_height=50)
 lrp_plot.scatter('time','lrptime', marker='circle', size=5, alpha=0.5, source=lrptime_source)
 lrp_plot.xaxis.axis_label = 'Time'
-lrp_plot.yaxis.axis_label = 'Total normalized LRP'
+lrp_plot.yaxis.axis_label = 'Normalized relevance score'
 
 
 #Layout