script file added

Notebooks/RESNET152/indiandanceprediction-usingmonk.py

 #!/usr/bin/env python
 # coding: utf-8
 
-# In[1]:
+# In[16]:
 
 
 pip install tornado --upgrade --use-feature=2020-resolver
 
 
-# In[2]:
+# In[17]:
 
 
 get_ipython().system('git clone https://github.com/Tessellate-Imaging/monk_v1.git')
 get_ipython().system('cd monk_v1/installation/Misc && pip install -r requirements_kaggle.txt --use-feature=2020-resolver')
 
 
-# In[3]:
+# In[ ]:
 
 
-pip install bokeh
+pip install bokeh --use-feature=2020-resolver
 
 
-# In[4]:
+# In[ ]:
 
 
-pip install monk --upgrade
+pip install monk --upgrade --use-feature=2020-resolver
 
 
-# In[5]:
+# In[ ]:
 
 
-pip install -U monk-kaggle
+pip install -U monk-kaggle --use-feature=2020-resolver
 
 
-# In[6]:
+# In[18]:
 
 
 import numpy as np # linear algebra
@@ -43,30 +43,29 @@ import os
 import sys
 
 
-# In[7]:
+# In[19]:
 
 
 sys.path.append("/kaggle/working/monk_v1/monk/");
 from gluon_prototype import prototype
 
 
-# In[8]:
+# In[20]:
 
 
 gtf = prototype(verbose=1);
 gtf.Prototype("Dance_Form", "Indian_Classical_Dance_Form_Prediction");
 
 
-# In[10]:
+# In[21]:
 
 
 import csv
 
 
-# In[11]:
+# In[22]:
 
 
-#to read the entries from train.csv
 data = []
 with open('../input/dform-gold/dataset1/train.csv', 'r',) as file:
     reader = csv.reader(file, delimiter = ',')
@@ -79,19 +78,18 @@ with open('../input/dform-gold/dataset1/train.csv', 'r',) as file:
 print(data)
 
 
-# In[12]:
+# In[23]:
 
 
-#used for cross validation, k : #iterations, div : #images in a division
 k = 4
 div = len(data) / k
+#7 * 13* 2 * 2
 
 
-# In[15]:
+# In[24]:
 
 
-#to shuffle data and get a set which classifies all 8 classes
-print("division:",div)
+print(div)
 
 #shuffling data
 
@@ -116,6 +114,8 @@ while (not flag):
         test = data[ i * int(div) : (i + 1) * int(div) ]
         for x in data[(i+1) * int(div) : ]:
             train.append(x)
+        print(i, "train count",len(pd.DataFrame(train).groupby(1).count()[0]))
+        print(i, "test count",len(pd.DataFrame(train).groupby(1).count()[0]))
         if(len(pd.DataFrame(train).groupby(1).count()[0]) < 8 or len(pd.DataFrame(test).groupby(1).count()[0]) < 8):
             flag = False
             break
@@ -124,28 +124,10 @@ while (not flag):
 print(data)
 
 
-# In[18]:
+# In[25]:
 
 
-#performed cross validation and created confusion matrix for cross validation
-'''
-    Model : resnet152_v1
-    #epochs:20
-    optimizer:adam
-    batch_size:7
-    learning_rate:0.005
-    data shuffle : true
-        
-'''
-'''
-    Result: 
-    Epochs for every iteration:
-    test accuracy, train accuracy, test loss, train loss
-    Summary of the model
-    Best accuracy after every iteration
-    Final average accuracy taken from the #iterations
-    
-'''
+print(data)
 accuracy = 0;
 for i in range(k):
     
@@ -163,20 +145,24 @@ for i in range(k):
     a.to_csv("test_slot.csv", index=False);
 
     gtf = prototype(verbose=1);
-    gtf.Prototype("Dance_Form", "resnet152_V1");
+    gtf.Prototype("Dance_Form", "Model");
 
 
 
-    gtf.Default(dataset_path=["../input/dform-gold/dataset1/train", "../input/dform-gold/dataset1/train"], path_to_csv=["train_slot.csv","test_slot.csv"], model_name="resnet152_v1", freeze_base_network=True, num_epochs=20);
-    gtf.update_shuffle_data(True);
-    gtf.optimizer_adam(0.001);
-    gtf.update_batch_size(7);
-    gtf.update_learning_rate(0.005);
+    gtf.Default(dataset_path=["../input/dform-gold/dataset1/train", "../input/dform-gold/dataset1/train"], path_to_csv=["train_slot.csv","test_slot.csv"], model_name="resnet152_v1", freeze_base_network=False, num_epochs=10);
+    #gtf.update_trainval_split(0.99);
+#     gtf.update_shuffle_data(True);
+#     gtf.optimizer_adam(0.001);
+#     gtf.update_batch_size(7);
+#     gtf.update_learning_rate(0.005);
 
     gtf.Reload()
 
     gtf.Train();
     
+    # gtf.Dataset_Params(dataset_path="../input/dform-gold/dataset1/test", path_to_csv="../input/dform-gold/dataset1/gold_labels.csv");
+    #list_test=os.listdir("../input/dform-gold/dataset1/test/");
+    #from tqdm.notebook import tqdm
     combined = [];
     predictions_dict = {};
     j = 0
@@ -197,7 +183,7 @@ for i in range(k):
     print(predictions_dict)
     print("predictions\n\n\n\n\n")
     print(combined)
-    gtf.Summary()
+    # gtf.Summary()
     gtf.EDA(show_img=True, save_img=True);
     # print(gtf.EDA(show_img=True, save_img=True))
     import csv
@@ -230,19 +216,14 @@ for i in range(k):
     print("correct",correct,"#images",div)
     print("accuracy for", i, "th iteration:", accuracy_of_iter)
     accuracy = accuracy + accuracy_of_iter
-    print(confusion_mat)
+#     print(confusion_mat)
 
 print("Overall accuracy:", accuracy / k)
 
 
-# In[19]:
+# In[26]:
 
 
-'''
-checking the model on test by comparing the gold labels of the test data and the prediction of model
-Creating confusion Matrix(real classes vs predicted ones)
-Calculating Final test accuracy
-'''
 confusion_mat = {
 
                      'bharatanatyam':{'bharatanatyam':0,'kathak':0,'kathakali':0,'kuchipudi':0,'manipuri':0,'mohiniyattam':0,'odissi':0,'sattriya':0}, 
@@ -282,5 +263,5 @@ print(conf_mat)
 # In[ ]:
 
 
-
+gtf.List_Models()