Handle activations in saved file, handle args in run.py

scripts/activations.py

@@ -64,7 +64,7 @@ def arctan_prime(y) -> np.ndarray:
     return 1 / y**2 + 1
 
 
-def listToActivations(activations_list, architecture):
+def listToActivations(activations_list, architecture) -> list:
     activations_fn = []
     activations_prime = []
     for id, _ in enumerate(architecture):

scripts/loulou.py

@@ -121,14 +121,14 @@ def train(weights: list, trX: np.ndarray, trY: np.ndarray, teX: np.ndarray, teY:
             if save_timeout > 0:
                 if i % save_timeout == 0:
                     temp_filename = '../trains/temp/' + \
-                        filename + '_epoch_' + str(i) + '.npy'
+                        filename + '_epoch_' + str(i) + '.npz'
                     temp_filename = os.path.join(path, temp_filename)
                     utils.save(weights, activations_fn,
                                temp_filename, reduce_output)
 
     # Save final file
     if filename:
-        filename = os.path.join(path, '../trains/' + filename + '.npy')
+        filename = os.path.join(path, '../trains/' + filename + '.npz')
         utils.save(weights, activations_fn, filename, reduce_output)
 
     return accuracy

scripts/run.py

@@ -1,44 +1,67 @@
 #!/usr/bin/env python3
 
+import matplotlib.image as image
 import numpy as np
+import argparse
 import sys
-import matplotlib.image as image
 
 from loulou import feed_forward
 from utils import convertJson
+import activations
 
 if __name__ == '__main__':
-    #   Handling errors for bad arguments
-    try:
-        assert len(sys.argv) == 3
-    except AssertionError:
-        print("Error ! Please give two arguments : path to weights file and to image to predict.")
-        exit()
+    verbosity = 0
+
+    parser = argparse.ArgumentParser(
+        description='Utility to run a loulou-based neural network.')
+    parser.add_argument('-f', '--file', dest='file', type=str, required=True,
+                        help='path to the `.npz` training file')
+    parser.add_argument('-i', '--image', dest='image', type=str, required=True,
+                        help='path to the image to predict')
+    parser.add_argument('-v', '--verbosity', dest='verbosity', action="count",
+                        help='add verbosity to the output (you can type several)')
+    parser.add_argument('-j', '--return-json', dest='return_json', action="store_true",
+                        help='print the prediction and hot vector in a json format')
+    args = parser.parse_args()
 
-    #   Loading weights matrix
-    filename = sys.argv[1]
+    if args.verbosity is not None:
+        verbosity = args.verbosity
+
+    # Load weights and activations from file
     try:
-        file = np.load(filename)
+        # TODO change behavior to allow_pickle=False for security
+        file = np.load(args.file, allow_pickle=True)
         weights = file['weights']
-        activations = file['activations']
+        activations_names = file['activations']
+
+        activations_fn = activations.listToActivations(
+            activations_names, weights)[0]
+        if verbosity > 0:
+            print('Activations used :', activations_names)
+
     except FileNotFoundError:
         print(
-            "Error ! Weights matrix file could not be opened, please check that it exists.")
-        print("Fichier : ", filename)
+            'Error ! File ['+args.file+'] could not be opened, please check that it exists.')
         exit()
 
-    #   Loading image data
-    img = sys.argv[2]
+    # Load image data
     try:
-        img = image.imread(img)
+        img = image.imread(args.image)
     except FileNotFoundError:
-        print("Error ! Image could not be opened, please check that it exists.")
-        print("Image : ", img)
+        print(
+            'Error ! Image ['+args.image+'] could not be opened, please check that it exists.')
         exit()
 
-    #   Shaping image onto matrix
+    # Convert image to matrix
     topred = 1 - img.reshape(784, 4).mean(axis=1)
-    #   Making prediction
-    prediction = feed_forward(topred, weights, activations)[-1]
-    #   Printing json output
-    print(convertJson(prediction))
+    # Make the prediction
+    prediction = feed_forward(topred, weights, activations_fn)[-1]
+    # Print final output
+    if args.return_json:
+        print(convertJson(prediction))
+    else:
+        if verbosity > 0:
+            print('Hot ones vector :', list(prediction))
+            print('Final prediction :', prediction.argmax())
+        else:
+            print(prediction.argmax())

scripts/train.py

@@ -21,8 +21,9 @@
     #   Parser
     parser = argparse.ArgumentParser(
         description='Utility to train a loulou-based neural network.')
+    # TODO set option to choose path
     parser.add_argument('-f', '--filename', dest='filename', type=str,
-                        help='name of the file to write, extension added automatically (default : none)')  # Need to set option for other paths
+                        help='name of the file to write, extension added automatically (default : none)')
     parser.add_argument('-e', '--epochs', dest='epochs', type=int,
                         help='number of epochs (default : 15, -1 for infinity)')
     parser.add_argument('-b', '--batch-size', dest='batch_size', type=int,
@@ -38,7 +39,7 @@
     parser.add_argument('-r', '--reduce-output', dest='reduce_output', action="count",
                         help='reduce verbosity of output (you can type several)')
     parser.add_argument('-j', '--return-json', dest='return_json', action="store_true",
-                        help='finally, print the progression of accuracy in a json format')
+                        help='print the progression of accuracy in a json format')
     args = parser.parse_args()
 
     if args.epochs is not None:

scripts/utils.py

@@ -2,17 +2,21 @@
 import json
 
 
-def save(weights: list, activations_list: list, filename: str, reduce_output: int) -> None:
-    np.save(filename, {'weights': weights,
-                       'activations': activations_list})
+def save(weights: list, activations_fn_list: list, filename: str, reduce_output: int) -> None:
+    activations_names_list = []
+    for act in activations_fn_list:
+        activations_names_list.append(act.__name__)
+
+    np.savez_compressed(filename, weights=weights,
+                        activations=activations_names_list)
     if reduce_output < 2:
         print('Data saved successfully into ', filename)
 
 
 def convertJson(pred: np.ndarray) -> str:
     out = {}
     out['hot_prediction'] = list(pred)
-    out['prediction'] = int(np.argmax(pred))
+    out['prediction'] = int(pred.argmax())
     return json.dumps(out)
 
 

todo.md

@@ -1,8 +1,6 @@
 # To-do
 
 - Update README
-- Handle args in `run.py` just like in `train.py`
 - Write descriptions for functions
 - Handle others types of data
-- Save trainings to an archive with datas : `weights matrix`, `activations arch`, [`hyperparameters`], [`accuracy`]
 - Rewrite as object ? (OOP)