Initial commit

Signed-off-by: Philip Abbet <philip.abbet@idiap.ch>

README.txt

@@ -0,0 +1,114 @@
+############################################################################
+
+             Emotion-Based Recommendation Generator (EMORec v1.0)      
+
+############################################################################
+
+
+README:
+=======
+A Python library which performs emotion-based analysis and recommendation using a 
+multiple-instance regression algorithm for a set of multimedia items described by 
+transcripts. The algorithm is trained over 1200 TED talks using the original human-
+made transcripts and the corresponding community emotion labels. The library can be 
+used in command line or directly in a Python program. It takes as input a JSON file 
+which contains an array of dictionaries that describe the metadata of multimedia items 
+and generates an output JSON file which contains the same items augmented with the 
+following attributes:
+
+    emotion_classes         The class names of 12 TED community emotion labels
+    emotion_scores          Estimated values for 12 TED community emotion labels		
+    emotion_rec             Recommended items based on these emotions	
+    emotion_rec_scores      Confidence of the recommended item
+    emotion_segments        Textual segments that were used
+        text                The actual textual content of the segment
+        start_time          Starting time of the segment
+        end_time            Ending time of the segment
+        relevance_scores    Relevance which reveals the contribution of the segment 
+                            to the prediction of the 14 emotion dimensions.
+
+FILES:
+======
+The library contains the following files:
+
+    ap_weights.py     Data class for items (text extraction, preprocessing)
+    crls.py           Vector space class supporting TF-IDF, LSI, RP and LDA
+    generate.py       Main class responsible for generating recommendations
+    data/             Data to be used for training
+    models/           Pre-trained regression models on TED for emotion prediction
+    parameters/       Optimal values obtained from cross-validation to be used
+                      for training and prediction
+
+
+USAGE:
+======
+USAGE: python generate.py -input=<path> -output=<path>
+	-input	 Path location of the input file in JSON format
+	-output	 Path location of the output file in JSON format
+
+EXAMPLE:
+========
+$  python generate.py --input=input.json --output=output.json --debug
+{'--debug': True,
+ '--display': False,
+ '--help': False,
+ '--input': 'input.json',
+ '--output': 'output.json',
+ '--version': False}
+[+] Loading items:....................................[OK]
+[+] Modeling emotions:
+        -> Unconvincing...............................[OK]
+        -> Fascinating................................[OK]
+        -> Persuasive.................................[OK]
+        -> Ingenious..................................[OK]
+        -> Longwinded.................................[OK]
+        -> Funny......................................[OK]
+        -> Inspiring..................................[OK]
+        -> Jaw-dropping...............................[OK]
+        -> Courageous.................................[OK]
+        -> Beautiful..................................[OK]
+        -> Confusing..................................[OK]
+        -> Obnoxious..................................[OK]
+[+] Generating recommendations........................[OK]
+[+] Saving to output file.............................[OK]
+[x] Finished.
+
+DEPENDENCIES:
+============
+1) Install python: http://www.python.org/getit/
+2) Install pip: http://www.pip-installer.org/en/latest/installing.html
+3) Then:
+$ pip install docopt
+$ pip install json
+$ pip install pyyaml
+$ pip install numpy
+$ pip install scipy
+$ pip install gensim
+$ pip install nltk
+$ python
+>>> import nltk
+>>> nltk.download()
+
+TROUBLESHOOTING:
+================ 
+Q: How can I use the library with items stored in other formats than JSON?
+A: You have to convert your file to JSON.
+Q: How can I use the library directly inside a Python program?
+A: Simply import the library in Python and initialize a generator object with 
+   the item dictionary of your preference.
+Q: Is there any attribute that is required to be present in the item metadata?
+A: Yes the 'id' attribute is mandatory.
+
+CONTACT:
+========
+Nikolaos Pappas 
+Idiap Research Institute
+Centre du Parc, 
+CH 1920 Martigny, 
+Switzerland
+E-mail:  nikolaos.pappas@idiap.ch 
+Website: http://people.idiap.ch/npappas/ 
+
+---
+Last update:
+8 Jul, 2014

ap_weights.py

@@ -0,0 +1,120 @@
+#    Copyright (c) 2014 Idiap Research Institute, http://www.idiap.ch/
+#    Written by Nikolaos Pappas <nikolaos.pappas@idiap.ch>,
+#
+#    This file is part of EMORec.
+#
+#    EMORec is free software: you can redistribute it and/or modify
+#    it under the terms of the GNU General Public License version 3 as
+#    published by the Free Software Foundation.
+#
+#    EMORec is distributed in the hope that it will be useful,
+#    but WITHOUT ANY WARRANTY; without even the implied warranty of
+#    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#    GNU General Public License for more details.
+#
+#    You should have received a copy of the GNU General Public License
+#    along with Foobar. If not, see <http://www.gnu.org/licenses/>.
+
+import numpy as np
+from scipy import sparse
+from scipy import linalg 
+from numpy import dot, matrix
+from sklearn import linear_model
+from numpy.linalg import norm
+from scipy.sparse import lil_matrix
+from sklearn.base import BaseEstimator
+from constrained_rls import cRLS 
+from sklearn.metrics import mean_absolute_error	
+
+
+class APWeights(BaseEstimator):
+
+	def __init__(self, iterations, l1=.5, l2=.5, l3=10, reg=None):
+		self.iterations = iterations 
+		self.l1 = l1					
+		self.l2 = l2  					
+		self.l3 = l3					
+		self.f1 = None
+		self.f2 = None
+		self.f3 = None  
+
+
+	def fit(self, X, Y):
+		M = X[0].get_shape()[1]      # number of features
+		N = len(X)                   # number of instances 
+		F = np.random.ranf((1,M))    # hyperplane to be learned
+ 		H = matrix(np.zeros((N,M)))  # bag representations
+ 		P = []
+		Y_w = []
+		X_w = []
+		converged = False
+		prev_error = 999999
+		it = 0
+		print "-"*100
+
+		print "L1: %f" % self.l1
+		print "L2: %f" % self.l2
+		print "L3: %f" % self.l3
+		print "M: %d" % M
+		print "N: %d" % N
+
+		print
+		print "[+] Training..." 
+		while(not converged and it < self.iterations):
+			for i, Xi in enumerate(X): 
+				if it == 0:
+					if X_w == []:
+						X_w = Xi
+					else:						
+						X_w = sparse.vstack([X_w, Xi ]) 
+					P.append(np.ones((1,X[i].get_shape()[0]))) 
+					Y_w.append([])
+				Xi = Xi.tocsr()
+				if self.f2: 
+					HC = matrix(self.f2.predict(Xi)).T 
+				else:
+					HC = Xi.dot(F.T).T
+				self.f1 = cRLS(alpha=self.l1)
+				P[i] = self.f1.fit(HC,Y[i],P[i])  
+				Y_w[i] = self.f1.coef_
+				cur_p = sparse.csr_matrix(self.f1.coef_)
+				H[i] = cur_p.dot(Xi).todense()
+
+			self.f2 = linear_model.Ridge(alpha=self.l2)
+			self.f2.fit(H,Y) 
+			pred = self.f2.predict(H)
+			cur_error = mean_absolute_error(pred,Y)
+			print "iteration %d -> (MAE: %f) " % (it, cur_error)
+			self.coef_ = self.f2.coef_
+			if prev_error - cur_error < 0.000001:
+				converged = True
+				self.coef_ = self.f2.coef_
+			prev_error = cur_error
+			it += 1 
+		Y_w = np.hstack(Y_w) 
+		print "Training f3..."
+		self.f3 = linear_model.Ridge(alpha=self.l3)  
+		self.f3.fit(X_w,Y_w)		
+
+		self.P = P
+		self.H = H
+		print "--/end"
+
+		return F
+
+
+	def predict(self, X):
+		M = X[0].get_shape()[1]       
+		N = len(X)				   	
+		H = matrix(np.zeros((N,M)))
+ 		W = []
+		for i, instances in enumerate(X):
+			Xi = instances 
+			weights = matrix(self.f3.predict(Xi)).view(np.ndarray)[0]
+			nweights = weights/sum(weights)
+			W.append(nweights)
+			H[i] =  dot(weights, Xi.todense())[0] 
+		Y = self.f2.predict(H)
+		self.P_test = W
+		return Y
+