validating train - kamangir/bolt#689

image_classifier/__init__.py

@@ -1,5 +1,5 @@
 name = "image_classifier"
 
-version = "1.1.44"
+version = "1.1.45"
 
 description = "fashion-mnist + hugging-face + awesome-bash-cli"

image_classifier/classes.py

@@ -1,8 +1,12 @@
 from .plot import *
 from abcli import file
+from abcli import path
 from abcli import string
 import numpy as np
 import matplotlib.pyplot as plt
+import tensorflow as tf
+from tqdm import *
+import time
 import abcli.logging
 import logging
 
@@ -60,9 +64,7 @@ class Image_Classifier(object):
 
         return True
 
-    def predict(self, test_images, test_labels, output_path="", options=""):
-        options = Options(options).default("cache", False).default("page_count", -1)
-
+    def predict(self, test_images, test_labels, output_path="", page_count=-1):
         logger.info(
             "image_classifier.predict({},{}){}".format(
                 string.pretty_size_of_matrix(test_images),
@@ -100,10 +102,6 @@ class Image_Classifier(object):
             cm = cm / np.sum(cm, axis=1)[:, np.newaxis]
             logger.debug("confusion_matrix: {}".format(cm))
 
-            if options["cache"]:
-                if not cache.write("{}.confusion_matrix".format(self.object_name), cm):
-                    return False
-
             if not file.save("{}/confusion_matrix.pyndarray".format(output_path), cm):
                 return False
 
@@ -147,8 +145,8 @@ class Image_Classifier(object):
                 return False
 
         max_index = test_images.shape[0]
-        if options["page_count"] != -1:
-            max_index = min(24 * options["page_count"], max_index)
+        if page_count != -1:
+            max_index = min(24 * page_count, max_index)
         offset = int(np.max(np.array(objects.list_of_frames(output_path) + [-1]))) + 1
         logger.info(
             "image_classifier.predict(offset={}): rendering {} frame(s)...".format(
@@ -285,16 +283,9 @@ class Image_Classifier(object):
         ]
 
     @staticmethod
-    def train(data_path, model_path, options=""):
-        options = (
-            Options(options)
-            .default("color", False)
-            .default("convnet", True)
-            .default("epochs", 10)
-        )
-
-        classifier = image_classifier()
-        classifier.params["convnet"] = options["convnet"]
+    def train(data_path, model_path, color=False, convnet=True, epochs=10):
+        classifier = Image_Classifier()
+        classifier.params["convnet"] = convnet
 
         logger.info(
             "image_classifier.train({}) -{}> {}".format(
@@ -304,7 +295,7 @@ class Image_Classifier(object):
             )
         )
 
-        success, train_images = file.load("{}/train_images.pyndarray".format(data_path))
+        success, train_images = file.load(f"{data_path}/train_images.pyndarray")
         if success:
             success, train_labels = file.load(f"{data_path}/train_labels.pyndarray")
         if success:
@@ -326,14 +317,14 @@ class Image_Classifier(object):
         window_size = train_images.shape[1]
         input_shape = (
             (window_size, window_size, 3)
-            if options["color"]
+            if color
             else (window_size, window_size, 1)
-            if options["convnet"]
+            if convnet
             else (window_size, window_size)
         )
         logger.info(f"input_shape:{string.pretty_size(input_shape)}")
 
-        if options["convnet"] and not options["color"]:
+        if convnet and not color:
             train_images = np.expand_dims(train_images, axis=3)
             test_images = np.expand_dims(test_images, axis=3)
 
@@ -343,15 +334,13 @@ class Image_Classifier(object):
         ):
             logger.info("{}: {}".format(name, string.pretty_size_of_matrix(thing)))
         logger.info(
-            "{} class(es): {}".format(
-                len(classifier.class_names), classifier.class_names
-            )
+            f"{len(classifier.class_names)} class(es): {', '.join(classifier.class_names)}"
         )
 
         train_images = train_images / 255.0
         test_images = test_images / 255.0
 
-        if options["convnet"]:
+        if convnet:
             # https://medium.com/swlh/convolutional-neural-networks-for-multiclass-image-classification-a-beginners-guide-to-6dbc09fabbd
             classifier.model = tf.keras.Sequential(
                 [
@@ -396,7 +385,7 @@ class Image_Classifier(object):
             metrics=["accuracy"],
         )
 
-        classifier.model.fit(train_images, train_labels, epochs=options["epochs"])
+        classifier.model.fit(train_images, train_labels, epochs=epochs)
 
         test_accuracy = float(
             classifier.model.evaluate(test_images, test_labels, verbose=2)[1]