{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"Deep Learning con Keras - Lezione 1","provenance":[],"collapsed_sections":["PaTwzBNg_ehy","Mt8hlgCKAlQP","yUMIUEOq_Odr","wPbNvmAZGHj_","DCcTH7De7Dva"],"toc_visible":true,"authorship_tag":"ABX9TyOfG25I+OULfk/YAbuaXc8C"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"}},"cells":[{"cell_type":"markdown","source":["# Introduzione\n","\n","**Deep Learning con Keras** è un breve ciclo di lezioni sull'utilizzo della libreria Keras per il deep learning.\n","\n","Le lezioni sono basate sul libro\n","\n","> FRANÇOIS CHOLLET
\n","> Deep Learning with Python
\n","> Manning\n","\n","e saranno tenute da\n","\n","> prof. [Gianluca Amato](https://www.sci.unich.it/~amato)
\n","> Dipartimento di Economia
\n","> Università \"G. d'Annunzio\" di Chieti-Pescara\n"],"metadata":{"id":"PaTwzBNg_ehy"}},{"cell_type":"markdown","source":["## Cosa è il deep learning ?\n","\n","Il deep learning (apprendimento profondo) è un settore del machine learning (apprendimento automatica), che è a sua volta una branca dell'intelligenza artificiale.\n","\n","![](\"https://drive.google.com/uc?export=view&id=1SEqW5lNd7pWGhXkNMtBXtsF675mZyRUQ\")
\n","\n","Le reti neurali artificiali (artificial neural networks) sono una tecnica di machine learning ispirata dal funzionamento del cervello umano."],"metadata":{"id":"Mt8hlgCKAlQP"}},{"cell_type":"markdown","source":["## Caso di studio: riconoscimento del contenuto di una immagine\n","\n","Riconoscimento di cifre (da 0 a 9) scritte a mano. Abbiamo a disposizione 70.000 immagini prese dal database MNIST (Modified National Institute of Standards and Technology), e vogliamo un programma che, presa una immagine, ci restituisca un intero tra 0 e 9 che è la cifra rappresentata in essa. \n","\n","Questo è un campione delle immagini che abbiamo a disposizione:\n","\n","![](\"https://drive.google.com/uc?export=view&id=1pK0Mm85Oj_rwOWlwhf92G-hcR8BM6Dj7\")
\n","\n","È un compito banale per gli esseri umani, ma se provassimo a scrivere noi stessi un algoritmo per il riconoscimento di queste immagini, ci accorgeremmo subito che non è per niente banale. La semplice intuizione di come riconoscere le varie forme (come \"un 9 ha un cappio in cima e un tratto verticale in basso a destra\") non è facile da esprimere algoritmicamente. Se si prova a rendere questa regole intuitive più precise, ci si ritrova persi in una miriade di eccezioni e casi particolari."],"metadata":{"id":"yUMIUEOq_Odr"}},{"cell_type":"markdown","source":["## Come ci aiuta il machine learning\n","\n","Con il machine learning noi non scriviamo un algoritmo specifico di riconoscimento delle immagini, ma utilizziamo degli algoritmi generici che \"imparano\" dai dati.\n","\n","![](\"https://drive.google.com/uc?export=view&id=1oIqsP5XYiamhbN9VqilxoGYSOzjLSSBL\")
\n","\n","Un algoritmo di *apprendimento* prende in input le immagini dell'insieme di addestramento, ognuna accompagnata dalla etichetta che indica di quale cifra si tratta, e genera un modello di classificazione (nel nostro caso una rete neurale, ma esistono altre possibilità).\n","\n","Un algoritmo di *inferenza* prende in input la rete neurale generale dalla fase di apprendimento e nuove immagini, assegnando ad ognuna di queste una etichetta da a 0 a 9. Ovviamente, a seconda di quanto è buono il modello ottenuto durante l'addestramento, i risultati dell'infernza possono essere più o meno corretti."],"metadata":{"id":"wPbNvmAZGHj_"}},{"cell_type":"markdown","source":["# Ambiente di lavoro\n","\n","**Python**: è il linguaggio per eccellenza nelle applicazioni di machine learning. Sebbene concettualmente sia un linguaggio imperativo ad oggetti come Java, la sintassi è abbastanza diversa. \n","\n","**Notebook**: molto spesso nell'ambito di deep learning non si scrivono programmi completi ma degli spezzoni di codice che vengono eseguiti in maniera interattiva. Il notebook è uno degli ambienti più comodi per l'uso interattivo.\n","\n","Un notebook è costituito da un insieme di celle. Le celle di testo, come questa che state leggendo, contiene del testo qualunque destinato ad esseree letto da un essere umano. Le celle di codice contengono invece del codice Python da eseguire.\n","\n","Ad esempio, in una cella di codice posso scrivere una espressione arimetica. Il risultato viene visualizzato immediatamente sotto."],"metadata":{"id":"MwyQ2M6_OCm5"}},{"cell_type":"code","source":["4+5*9"],"metadata":{"id":"UlNvxD-o_DAP","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947728610,"user_tz":-60,"elapsed":6,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"548fb813-a79a-4f58-f097-3052963bdd9d"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["49"]},"metadata":{},"execution_count":1}]},{"cell_type":"markdown","source":["Se voglio usare delle funzioni predefinite di Python, può essere necessario importarle. Ad esempio, la funzione radice quadrata (`sqrt`) sta nel modulo `math`, ma non si può usare immediatamente."],"metadata":{"id":"DIGfrDpl_UmI"}},{"cell_type":"code","source":["math.sqrt(2)"],"metadata":{"id":"h5zR7c0_AUCB","colab":{"base_uri":"https://localhost:8080/","height":167},"executionInfo":{"status":"error","timestamp":1647947730586,"user_tz":-60,"elapsed":7,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"59f33c5f-709f-4051-f4e4-ba43ad7c3123"},"execution_count":null,"outputs":[{"output_type":"error","ename":"NameError","evalue":"ignored","traceback":["\u001b[0;31m---------------------------------------------------------------------------\u001b[0m","\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)","\u001b[0;32m\u001b[0m in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msqrt\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m","\u001b[0;31mNameError\u001b[0m: name 'math' is not defined"]}]},{"cell_type":"markdown","source":["Prima bisogna importare il modulo `math` con il comando `import`.\n","\n"],"metadata":{"id":"Is8VONT3Alkb"}},{"cell_type":"code","source":["import math\n","\n","math.sqrt(2)"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"ZQw3ilpD_dfp","executionInfo":{"status":"ok","timestamp":1647947733432,"user_tz":-60,"elapsed":5,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"c83966eb-6d9b-41f4-fb79-831c6724d80b"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["1.4142135623730951"]},"metadata":{},"execution_count":3}]},{"cell_type":"markdown","source":["E a quel punto lo si può usare in tutte le celle, sia se vengono prima sia se vengono dopo il punto di importazione."],"metadata":{"id":"OghypmvRDpZR"}},{"cell_type":"code","source":["math.sqrt(2)"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"QQeDwuiID3kZ","executionInfo":{"status":"ok","timestamp":1647947736891,"user_tz":-60,"elapsed":333,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"4d92705f-ad48-4d56-df40-37c3c9d2d51c"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["1.4142135623730951"]},"metadata":{},"execution_count":4}]},{"cell_type":"markdown","source":["In una cella è possibile inserire un codice Python più lungo."],"metadata":{"id":"RO4Zqbk_D7L3"}},{"cell_type":"code","source":["# Esempio di codice Python che esegue la somma dei numeri da 0 fino ad n.\n","n = 10\n","sum = 0\n","for y in range(n+1):\n"," sum = sum + y\n"," print(y)\n","print(f\"La somma dei primi {n} numeri è {sum}\")"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"P19xEiOzVb4R","executionInfo":{"status":"ok","timestamp":1647947738768,"user_tz":-60,"elapsed":3,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"f2adcf98-d6f0-40c6-d755-d84abddce19f"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["0\n","1\n","2\n","3\n","4\n","5\n","6\n","7\n","8\n","9\n","10\n","La somma dei primi 10 numeri è 55\n"]}]},{"cell_type":"markdown","source":["La cella del notebook qui sopra **due delle caratteristiche di Python** che lo distinguono in maniera più evidente da Java:\n","\n","* le variabili non devono essere dichiarate prima di essere utilizzate\n","* l'indentazione sostituisce le parentesi graffe nello specificare blocchi di codice\n","\n","Una volta che un cella di codice è stata eseguita, le variabili create sono accessibili anche dalle altre celle\n"],"metadata":{"id":"wokRRYRlWPeB"}},{"cell_type":"code","source":["sum"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"d5ljnXbYXHbb","executionInfo":{"status":"ok","timestamp":1647947741867,"user_tz":-60,"elapsed":264,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"246173be-7055-443b-be03-837b2823375b"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["55"]},"metadata":{},"execution_count":6}]},{"cell_type":"markdown","source":["Il notebook può anche contenere output di tipo grafico"],"metadata":{"id":"Hrs2Bc-CXkMQ"}},{"cell_type":"code","source":["import matplotlib.pyplot as plt\n","\n","plt.plot([1, 2, 3, 4], [1, 4, 9, 16], 'ro')\n","plt.axis([0, 6, 0, 20])\n","plt.show()"],"metadata":{"colab":{"base_uri":"https://localhost:8080/","height":269},"id":"Vpio5R_RVem7","executionInfo":{"status":"ok","timestamp":1647947744182,"user_tz":-60,"elapsed":882,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"bb0dcfe7-c018-4059-fd1c-a6674ea86206"},"execution_count":null,"outputs":[{"output_type":"display_data","data":{"text/plain":[""],"image/png":"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\n"},"metadata":{"needs_background":"light"}}]},{"cell_type":"markdown","source":["È possibile ovviamente lavorare interamente nel proprio computer installando Python e Jupyter Notebook, ma per queste lezioni ho preferito utilizzare Google Colab."],"metadata":{"id":"b1oczSVRYFI4"}},{"cell_type":"markdown","source":["# Primi passi"],"metadata":{"id":"ndHn_zYHUrYW"}},{"cell_type":"code","source":["# Importiamo le librerie Numpy, TensorFlow e Matplotlib (quest'ultima per la visualizzazione grafica)\n","# Notare l'uso della clausola \"as\" per poter usare quel modulo con un nome più corto.\n","# Le abbreviazioni np, tf e plt sono una specie di standard.\n","\n","import numpy as np\n","import tensorflow as tf\n","import matplotlib.pyplot as plt"],"metadata":{"id":"0pMc7ij4Y4q4"},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":["## Caricamento dei dati di addestramento"],"metadata":{"id":"Y1ZZvgWsmCvh"}},{"cell_type":"code","source":["# Carichiamo il dataset MNIST\n","# - train_images sono le immagini per la fase di addestramento\n","# - train_labels sono le etichette associate a queste immagini (numeri da 0 a 9) \n","\n","# test_images e test_labels sono simili a train_images e train_labels ma costituiscono \n","# i dati per la fase di test (ne parleremo dopo)\n","\n","(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()"],"metadata":{"id":"R_8lwtovTtrE","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947775151,"user_tz":-60,"elapsed":625,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"fc742598-d45c-4e4c-dd6f-b95cfb867b14"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n","11493376/11490434 [==============================] - 0s 0us/step\n","11501568/11490434 [==============================] - 0s 0us/step\n"]}]},{"cell_type":"code","source":["# Proviamo a vedere il contenuto di train_images... non si capisce molto.\n","# Quello che si capisce è che si tratta di un array contenente numeri interi a \n","# 8 bit senza segno (vedi dtype=uint8 alla fine dell'output)\n","\n","train_images"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"u05hqo8wojAe","executionInfo":{"status":"ok","timestamp":1647947777581,"user_tz":-60,"elapsed":6,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"27bbe951-fb34-4217-e801-581694e4505a"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]],\n","\n"," [[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]],\n","\n"," [[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]],\n","\n"," ...,\n","\n"," [[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]],\n","\n"," [[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]],\n","\n"," [[0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," ...,\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0],\n"," [0, 0, 0, ..., 0, 0, 0]]], dtype=uint8)"]},"metadata":{},"execution_count":10}]},{"cell_type":"code","source":["# La funzione len ci restituisce il numero di elementi di un vettore.\n","# Nel nostro caso è 60.000 (il numero di immagini di addestramento).\n","\n","len(train_images)"],"metadata":{"id":"F-z30WfeUFch","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947792030,"user_tz":-60,"elapsed":424,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"b7507fb1-3e77-4a8a-ab84-7a7e3ea98fe7"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["60000"]},"metadata":{},"execution_count":11}]},{"cell_type":"code","source":["# Ogni immagine è una matrice di 28 x 28 elementi. Ogni elemento corrisponde ad un pixel dell'immagine.\n","# Il valore numerico è la gradazione di grigio di quel punto da 0 (bianco) a 255 (nero)\n","\n","train_images[2]"],"metadata":{"id":"7v4ROJ9FUHoe","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947869905,"user_tz":-60,"elapsed":240,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"1e0bb593-3b3c-47c4-f2b3-e4776bd04636"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 67, 232, 39, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 62, 81, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 120, 180, 39, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 126, 163, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 2, 153, 210, 40, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 220, 163, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 27, 254, 162, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 222, 163, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 183, 254, 125, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 46, 245, 163, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 198, 254, 56, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 120, 254, 163, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 23, 231, 254, 29, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 159, 254, 120, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 163, 254, 216, 16, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 159, 254, 67, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 14, 86, 178, 248, 254, 91, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 159, 254, 85, 0, 0, 0, 47, 49, 116, 144,\n"," 150, 241, 243, 234, 179, 241, 252, 40, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 150, 253, 237, 207, 207, 207, 253, 254, 250, 240,\n"," 198, 143, 91, 28, 5, 233, 250, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 119, 177, 177, 177, 177, 177, 98, 56, 0,\n"," 0, 0, 0, 0, 102, 254, 220, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 254, 137, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 254, 57, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 254, 57, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 255, 94, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 254, 96, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 254, 153, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 169, 255, 153, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 96, 254, 153, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0],\n"," [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n"," 0, 0]], dtype=uint8)"]},"metadata":{},"execution_count":12}]},{"cell_type":"code","source":["# complessivamente siamo di fronte ad un array di 60.000 x 28 x 28 interi.\n","# Le dimensioni complessive dell'array si possono esaminare con la proprietà\n","# shape. Questi vettori a 3-dimensioni (o anch più) si chiamano tensori\n","# nella terminologia di TensorFlow.\n","\n","train_images.shape"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"SwbCJCKBoEO0","executionInfo":{"status":"ok","timestamp":1647947873300,"user_tz":-60,"elapsed":261,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"fd5ed4e5-b5ee-4744-ab0d-80fe624ce766"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(60000, 28, 28)"]},"metadata":{},"execution_count":13}]},{"cell_type":"code","source":["# Ovviamente, se chiedo lo shape di una singola immagine, quella sarà\n","# solo 28 x 28.\n","\n","train_images[2].shape"],"metadata":{"id":"ufpetO-xIqKQ","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947915743,"user_tz":-60,"elapsed":240,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"7945334d-49bf-4533-9a7d-355bed135fa4"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(28, 28)"]},"metadata":{},"execution_count":15}]},{"cell_type":"code","source":["# Possiamo vedere il contenuto di una immagine in forma grafica col il comando plt.imshow\n","\n","plt.imshow(train_images[2], cmap = plt.cm.binary)"],"metadata":{"id":"C30mjMwWXy0w","colab":{"base_uri":"https://localhost:8080/","height":282},"executionInfo":{"status":"ok","timestamp":1647947918110,"user_tz":-60,"elapsed":241,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"0d01a4ea-f006-4eb4-be42-3663b4ab86c1"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":[""]},"metadata":{},"execution_count":16},{"output_type":"display_data","data":{"text/plain":[""],"image/png":"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\n"},"metadata":{"needs_background":"light"}}]},{"cell_type":"code","source":["# train_labels è un elenco di 60.000 valori da 0 a 9. Ogni valore è la cifra\n","# rappresentata nella corrispondente immagine. Si, noti, ad esempio, che\n","# train_labels[2] è 4, che è il numero rappresentato in train_imges[2].\n","\n","train_labels"],"metadata":{"id":"nkfWoooyapTK","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947927423,"user_tz":-60,"elapsed":225,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"fabfba94-cd6f-4c13-c770-6023b050d965"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([5, 0, 4, ..., 5, 6, 8], dtype=uint8)"]},"metadata":{},"execution_count":17}]},{"cell_type":"code","source":["train_labels[2]"],"metadata":{"id":"cB_S4tYoP-pP","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647947947841,"user_tz":-60,"elapsed":239,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"ff627bbc-b401-487d-e056-b3617fd916c0"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["4"]},"metadata":{},"execution_count":18}]},{"cell_type":"code","source":["# train_labels è un classico array ad una singola dimensione, lungo 60.000\n","\n","train_labels.shape"],"metadata":{"id":"mhTkSJypcg-l","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948051809,"user_tz":-60,"elapsed":241,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"dd9cbdf0-0b67-43e9-a024-3c0c14ed2989"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(60000,)"]},"metadata":{},"execution_count":19}]},{"cell_type":"markdown","source":["## Preparazione dei dati"],"metadata":{"id":"yzzKtrUih-b1"}},{"cell_type":"code","source":["# Occorre trasformare le etichette dal formato numerico originale ad un formato\n","# confrontabile con l'output del percettrone. Usiamo il metodo predefinito di\n","# tf.keras.utils.to_categorical.\n","\n","train_labels_ok = tf.keras.utils.to_categorical(train_labels)\n","test_labels_ok = tf.keras.utils.to_categorical(test_labels)"],"metadata":{"id":"UGt-pP9DgCyY"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["# Adesso al posto di ogni etichetta troviamo un vetore di 0 ed 1. Se l'etichetta\n","# è n, il corrispondente vettore è omposto da tutti 0 tranne un 1 in posizione n.\n","# Questa codifica prende il nome di One Hot Encoding.\n","\n","train_labels_ok"],"metadata":{"id":"2D8CID4dge56","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948112278,"user_tz":-60,"elapsed":242,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"40087d41-05a5-4759-a22f-178050bd21ba"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[0., 0., 0., ..., 0., 0., 0.],\n"," [1., 0., 0., ..., 0., 0., 0.],\n"," [0., 0., 0., ..., 0., 0., 0.],\n"," ...,\n"," [0., 0., 0., ..., 0., 0., 0.],\n"," [0., 0., 0., ..., 0., 0., 0.],\n"," [0., 0., 0., ..., 0., 1., 0.]], dtype=float32)"]},"metadata":{},"execution_count":21}]},{"cell_type":"code","source":["train_labels_ok[2]"],"metadata":{"id":"tiz39S3tJD2T","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948114855,"user_tz":-60,"elapsed":221,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"e054171b-8184-4435-a957-ef57d88382dc"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([0., 0., 0., 0., 1., 0., 0., 0., 0., 0.], dtype=float32)"]},"metadata":{},"execution_count":22}]},{"cell_type":"code","source":["# È prassi fornire i dati in input in maniera normalizzata, come numeri compresi\n","# tra 0 ed 1. Pertanto, anche se non strettamente necessario, dividiamo tutti i\n","# numeri che compongono una immagine per 255.\n","\n","train_images_ok = train_images / 255.0\n","test_images_ok = test_images / 255.0"],"metadata":{"id":"PU7lzXaZgxzI"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["train_images_ok[2]"],"metadata":{"id":"8gyltB2upvrT","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647183378834,"user_tz":-60,"elapsed":12,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"1e4b8ab8-5dab-4369-c5c5-f68eca0c59c3"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0.2627451 , 0.90980392, 0.15294118, 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0.24313725,\n"," 0.31764706, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0.47058824, 0.70588235, 0.15294118, 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0.49411765,\n"," 0.63921569, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0.00784314,\n"," 0.6 , 0.82352941, 0.15686275, 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0.8627451 ,\n"," 0.63921569, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0.10588235,\n"," 0.99607843, 0.63529412, 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0.87058824,\n"," 0.63921569, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0.71764706,\n"," 0.99607843, 0.49019608, 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.18039216, 0.96078431,\n"," 0.63921569, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0.77647059,\n"," 0.99607843, 0.21960784, 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.47058824, 0.99607843,\n"," 0.63921569, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0.09019608, 0.90588235,\n"," 0.99607843, 0.11372549, 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.62352941, 0.99607843,\n"," 0.47058824, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0.63921569, 0.99607843,\n"," 0.84705882, 0.0627451 , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.62352941, 0.99607843,\n"," 0.2627451 , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0.05490196, 0.3372549 , 0.69803922, 0.97254902, 0.99607843,\n"," 0.35686275, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.62352941, 0.99607843,\n"," 0.33333333, 0. , 0. , 0. , 0.18431373,\n"," 0.19215686, 0.45490196, 0.56470588, 0.58823529, 0.94509804,\n"," 0.95294118, 0.91764706, 0.70196078, 0.94509804, 0.98823529,\n"," 0.15686275, 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0.58823529, 0.99215686,\n"," 0.92941176, 0.81176471, 0.81176471, 0.81176471, 0.99215686,\n"," 0.99607843, 0.98039216, 0.94117647, 0.77647059, 0.56078431,\n"," 0.35686275, 0.10980392, 0.01960784, 0.91372549, 0.98039216,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0.46666667,\n"," 0.69411765, 0.69411765, 0.69411765, 0.69411765, 0.69411765,\n"," 0.38431373, 0.21960784, 0. , 0. , 0. ,\n"," 0. , 0. , 0.4 , 0.99607843, 0.8627451 ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 0.99607843, 0.5372549 ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 0.99607843, 0.22352941,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 0.99607843, 0.22352941,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 1. , 0.36862745,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 0.99607843, 0.37647059,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 0.99607843, 0.6 ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.6627451 , 1. , 0.6 ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0.37647059, 0.99607843, 0.6 ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ],\n"," [0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. , 0. , 0. ,\n"," 0. , 0. , 0. ]])"]},"metadata":{},"execution_count":47}]},{"cell_type":"code","source":["plt.imshow(train_images_ok[2], cmap= plt.cm.binary)"],"metadata":{"id":"l0XXFtNJpqNp","colab":{"base_uri":"https://localhost:8080/","height":282},"executionInfo":{"status":"ok","timestamp":1647948198901,"user_tz":-60,"elapsed":299,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"22206939-31fa-43e2-8958-73fbc7655ce2"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":[""]},"metadata":{},"execution_count":24},{"output_type":"display_data","data":{"text/plain":[""],"image/png":"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\n"},"metadata":{"needs_background":"light"}}]},{"cell_type":"markdown","source":["## Creazione della rete neurale"],"metadata":{"id":"QR_cty6BmHl-"}},{"cell_type":"code","source":["# Costruiamo una rete neurale simile al percettrone. Non è esattamente uguale perché la funzione\n","# di attivazione hard_sigmoid è simile ma non identica alla step function usata nel percettrone.\n","# Il primo strato (Flatten) ha lo copo di trasformare una immagine di 28 x 28 pixel in un unico vettore\n","# di 784 elementi, il secondo (Dense) è lo strato di neuroni. Ci sono 10 neuroni, ognuno connesso\n","# a tutti i 784 pixel dell'immagine.\n","\n","network = tf.keras.models.Sequential([\n"," tf.keras.layers.Flatten(input_shape=(28, 28)),\n"," tf.keras.layers.Dense(10, activation='hard_sigmoid')\n","])"],"metadata":{"id":"ijqRtHB-dbLc"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["# Visualizziamo la struttura della rete. Notate che la rete ha 7850 parametri. Ognuno dei 10 neuroni\n","# di output ha infatti 784 pesi (uno per ogni pixel) ed un valore di bias. In totale quindi sono\n","# (784 + 1) * 10 = 7850 pesi. Notare che lo strato Flatten eseguo un compito fisso e non ha\n","# nessun peso associato.\n","\n","network.summary()"],"metadata":{"id":"LJQ31oMQbUl2","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948227598,"user_tz":-60,"elapsed":269,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"cb0e611a-3c5b-45a9-87d4-d620b5399230"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Model: \"sequential\"\n","_________________________________________________________________\n"," Layer (type) Output Shape Param # \n","=================================================================\n"," flatten (Flatten) (None, 784) 0 \n"," \n"," dense (Dense) (None, 10) 7850 \n"," \n","=================================================================\n","Total params: 7,850\n","Trainable params: 7,850\n","Non-trainable params: 0\n","_________________________________________________________________\n"]}]},{"cell_type":"code","source":["# Prima di poter usare una rete, bisogna indicare dei parametri addizionali:\n","# - una funzione di loss, che specifica come valutare le differenze tra il vero valore degli\n","# output e quello predetto (mse = scarto quadrarico medio)\n","# - l'algoritmo di ottimizzazione che viene usato per aggiornare i pesi della rete \n","# nel tentativodi ridurre la funzione di loss (sgd = stochastic gradient descend)\n","# - metriche da visualizzare durante l'addestramento (in questo caso visualizziamo l'accuratezza, la\n","# percentuale di istanze classificate correttamente nell'insieme di addestramento)\n","\n","network.compile(optimizer='sgd', loss='mse', metrics=['accuracy'])"],"metadata":{"id":"ldsTGQkmef5M"},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":["## Addestramento rete e valutazione dei risultati"],"metadata":{"id":"eAUCCnwhpysw"}},{"cell_type":"code","source":["# Lanciamo l'addestramento della rete:\n","# - batch_size è il numero di istanze su cui addestrare i pesi as ogni passo\n","# - epochs è il numero di volte per cui ripetere l'addestramento sull'intero\n","# insieme di training (ogni volta usando come punto di partenza i pesi trovati\n","# nell'epoca precedente)\n","# Restituisce la storia dell'addestramento\n","\n","history = network.fit(train_images_ok, train_labels_ok, epochs=15, batch_size=128)"],"metadata":{"id":"Bf9mOwmyfc39","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948352201,"user_tz":-60,"elapsed":21409,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"777daef6-e91f-4074-84d4-cb20bee21552"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Epoch 1/15\n","469/469 [==============================] - 2s 2ms/step - loss: 0.1497 - accuracy: 0.0925\n","Epoch 2/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.1000 - accuracy: 0.2175\n","Epoch 3/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0902 - accuracy: 0.3668\n","Epoch 4/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0846 - accuracy: 0.4729\n","Epoch 5/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0801 - accuracy: 0.5420\n","Epoch 6/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0764 - accuracy: 0.5885\n","Epoch 7/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0732 - accuracy: 0.6234\n","Epoch 8/15\n","469/469 [==============================] - 1s 3ms/step - loss: 0.0703 - accuracy: 0.6522\n","Epoch 9/15\n","469/469 [==============================] - 2s 4ms/step - loss: 0.0678 - accuracy: 0.6752\n","Epoch 10/15\n","469/469 [==============================] - 2s 4ms/step - loss: 0.0655 - accuracy: 0.6971\n","Epoch 11/15\n","469/469 [==============================] - 2s 4ms/step - loss: 0.0635 - accuracy: 0.7163\n","Epoch 12/15\n","469/469 [==============================] - 2s 3ms/step - loss: 0.0617 - accuracy: 0.7329\n","Epoch 13/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0600 - accuracy: 0.7469\n","Epoch 14/15\n","469/469 [==============================] - 1s 3ms/step - loss: 0.0585 - accuracy: 0.7585\n","Epoch 15/15\n","469/469 [==============================] - 1s 2ms/step - loss: 0.0571 - accuracy: 0.7666\n"]}]},{"cell_type":"code","source":["# Calcoliamo le predizioni della rete neurale sull'insieme di addestramento\n","\n","train_predictions = network.predict(train_images_ok)"],"metadata":{"id":"6p-SJ2j3lHaB"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["# Vediamo ad esempio cosa produce la per l'immagine numero 2. Il valore di\n","# probabilità maggiore è il quinto (corrispondente alla cifra 4), per cui la\n","# nostra rete ha indovinato che l'immagine numero 2 è effettivamente un 4.\n","\n","train_predictions[2]"],"metadata":{"id":"uuYEbKHJPnAI","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948384088,"user_tz":-60,"elapsed":257,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"383f22ba-99b0-4505-b238-a6ae6ae44ce6"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([0.19969535, 0. , 0.3530274 , 0.27846837, 0.43973866,\n"," 0.23310304, 0.3095976 , 0.2790324 , 0.32828578, 0.3380107 ],\n"," dtype=float32)"]},"metadata":{},"execution_count":31}]},{"cell_type":"code","source":["# Per calcolare qual è l'indice col valore maggiore, possiamo usare la funzione np.argmax\n","\n","np.argmax(train_predictions[2])"],"metadata":{"id":"bwYBgEW-P-bW","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948387868,"user_tz":-60,"elapsed":273,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"8627b7df-b746-4640-c305-fbb718bd2c51"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["4"]},"metadata":{},"execution_count":32}]},{"cell_type":"code","source":["# Deteminiamo quali sono le immagini dove la nostra rete sbaglia\n","\n","errors = np.where([ np.argmax(x) for x in train_predictions] != train_labels)\n","errors"],"metadata":{"id":"KpZh8c27tFZ6","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948434645,"user_tz":-60,"elapsed":359,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"420b0514-a592-4fc6-bbeb-e922d1aac13b"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(array([ 11, 17, 19, ..., 59982, 59992, 59993]),)"]},"metadata":{},"execution_count":33}]},{"cell_type":"code","source":["# Controlliamo ad esempio l'immagine 19\n","\n","print(\"Predizione: \", np.argmax(train_predictions[19]))\n","print(\"Valore effettivo: \", train_labels[19])\n","\n","plt.imshow(train_images_ok[19], cmap= plt.cm.binary)"],"metadata":{"id":"7jk0s5Smw479","colab":{"base_uri":"https://localhost:8080/","height":317},"executionInfo":{"status":"ok","timestamp":1647948452054,"user_tz":-60,"elapsed":263,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"5f0d48ea-3311-473e-ee09-186999643b3a"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Predizione: 1\n","Valore effettivo: 9\n"]},{"output_type":"execute_result","data":{"text/plain":[""]},"metadata":{},"execution_count":35},{"output_type":"display_data","data":{"text/plain":[""],"image/png":"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\n"},"metadata":{"needs_background":"light"}}]},{"cell_type":"code","source":["train_predictions[19]"],"metadata":{"id":"jUMY8OB7T26Q","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948456327,"user_tz":-60,"elapsed":293,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"92769eeb-7b10-4bd9-a616-7dbaad3a1c35"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([0.2219345 , 0.36409962, 0.3848489 , 0.21659535, 0.22457659,\n"," 0.31142017, 0.23089674, 0.19564283, 0.3023529 , 0.16161236],\n"," dtype=float32)"]},"metadata":{},"execution_count":36}]},{"cell_type":"code","source":["# Valutiamo adesso il funzionamento della rete su immagini nuove, che non fanno\n","# parte dell'insieme di addestramento. Usiamo a tale scopo il metodo evaluate\n","# e l'insieme di test. Notiamo che l'accuratezza è simile a quella sull'insieme\n","# di addestramento, ma non sarà sempre così.\n","\n","network.evaluate(test_images_ok, test_labels_ok)"],"metadata":{"id":"aNpBfBmwk3Ee","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948489323,"user_tz":-60,"elapsed":840,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"c3ae7ef4-9876-45ac-db20-26e0a20b1b67"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["313/313 [==============================] - 1s 1ms/step - loss: 0.0558 - accuracy: 0.7791\n"]},{"output_type":"execute_result","data":{"text/plain":["[0.05584060028195381, 0.7791000008583069]"]},"metadata":{},"execution_count":37}]},{"cell_type":"code","source":["# Visualizziamo l'andamento dell'errore di addestramento\n","\n","history_dict = history.history\n","loss_values = history_dict['loss']\n","accuracy_values = history_dict['accuracy']\n","epochs = range(1, len(loss_values) + 1)\n","\n","plt.plot(epochs, loss_values, 'bo', label='Training loss')\n","plt.title('Training loss w.r.t epochs')\n","plt.xlabel('Epochs')\n","plt.ylabel('Loss')\n","plt.legend()\n","plt.show()\n","\n","plt.plot(epochs, accuracy_values, 'bo', label='Accuracy')\n","plt.title('Accuracy w.r.t. epochs')\n","plt.xlabel('Epochs')\n","plt.ylabel('Accuracy')\n","plt.legend()\n","plt.show()"],"metadata":{"id":"7ubrN-ehSgAt","colab":{"base_uri":"https://localhost:8080/","height":573},"executionInfo":{"status":"ok","timestamp":1647948493694,"user_tz":-60,"elapsed":776,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"1b4bb309-4bd9-4a32-9546-0e141d7ba947"},"execution_count":null,"outputs":[{"output_type":"display_data","data":{"text/plain":[""],"image/png":"iVBORw0KGgoAAAANSUhEUgAAAYgAAAEWCAYAAAB8LwAVAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAesUlEQVR4nO3deZgddZ3v8fcnnZCQhcUQWdJJOmgEA4SOdNiiGYSrBEHCKCiZvkJEDDgiiDgQyDggIzMuqFxGuEMjqzaCDyg3KsoiS0AQ6cSIBMIQMR0aWUICWQxLlu/9o6rDoVPdfTrp6uqT/ryep5+u+lXVOd9z0jmfU79fLYoIzMzM2upXdAFmZtY7OSDMzCyTA8LMzDI5IMzMLJMDwszMMjkgzMwskwPCegVJv5Z0cnev28UaDpPU0t2P25dJukjSj4uuw7ZM/6ILsMolaU3J7GDgTWBDOn9aRDSW+1gRcVQe6/YlkpYAp0bEPUXXYtsGB4RtsYgY2jrd0YeTpP4Rsb4na+sL/L5a3tzFZN2utatG0nmSXgSuk7SzpF9KWibp1XS6umSb+yWdmk7PkPSQpEvTdf8q6agtXHespLmSVku6R9IV5XZ5SHp/+lyvSVoo6diSZR+T9GT6uM9L+mravkv62l6TtELSg5I2+38m6euS/iudHiDp75K+k85vL+kNSe9qs81Fkm6V9GNJq4AZJct+BIwGfiFpjaRz23lNx0hakNb3sKQJJcuWSDo/fV2vSrpO0qCS5Z+XtDh9XXMk7VGybB9Jd6fLXpJ0QcnTbifpxvS9WiiprmS789L3b7WkpyUd0ek/jPUYB4TlZTfgXcAYYCbJ39p16fxo4HXgBx1sfxDwNLAL8G3gGknagnVvAv4ADAcuAj5TTvGSBgC/AO4C3g18CWiUtFe6yjUk3WjDgH2Be9P2c4AWYASwK3ABkHU9mweAw9LpScCLwJR0/hDg6YhYkbHdNOBWYCdgUxdeRHwGWAp8PCKGRsS3M17TROBa4DSS9+MqYI6kgSWr1QNHAu8B3gf8a7rt4cB/Ap8CdgeagZvTZcOAe4DfAHsA7wV+W/KYx6br7gTMIf13T9/LM4BJ6ft4JLAk4zVbQRwQlpeNwIUR8WZEvB4RyyPitohYGxGrgUuAf+hg++aIuDoiNgA3kHwo7dqVdSWNJvnw/beIeCsiHiL5gCrHwcBQ4JvptvcCvwSmp8vXAeMl7RARr0bE/JL23YExEbEuIh6M7AuePQKMkzScJBiuAUZKGkryvjzQTl2PRMTtEbExIl4v87W0mglcFRGPRsSGiLiBZNzo4JJ1fhARz6XhdEnJ660Hro2I+RHxJnA+cIikGuAY4MWI+G5EvBERqyPi0ZLHfCgi7kj/fX4E7J+2bwAGkryPAyJiSUT8pYuvyXLkgLC8LIuIN1pnJA2WdJWk5rR7ZC6wk6SqdrZ/sXUiItamk0O7uO4ewIqSNoDnyqx/D+C5iNhY0tYMjEynPwl8DGiW9ICkQ9L27wCLgbskPStpVtaDpx/uTSRhMIUkEB4GJtNxQJRbf5YxwDlp99Jrkl4DRpG81qzHby5Ztkc631r/GmA5yfsxCujog/3Fkum1wKB0/GQx8GWSPbuXJd1c2m1lxXNAWF7afms+B9gLOCgiduDt7pT2uo26wwvAuyQNLmkbVea2fwNGtRk/GA08DxARj0XENJLup9uBn6btqyPinIjYk6Rr5Ssd9Ks/ABwOTAQeS+ePBA4kCdAsHV1+ubNLMz8HXBIRO5X8DI6In5SsU/r+jCZ5H0h/j2ldIGkISTfV8+nj7tnJc2cXHHFTRHwwfewAvrUlj2P5cEBYTxlGMu7wWjr4emHeTxgRzSTf0i+StF36Lf/jZW7+KMm33XPTQeTD0m1vTh+rXtKOEbEOWEXSpdY6CPzedAxkJUk3ysbsp+AB4CTgyYh4C7gfOBX4a0Qs24KX/BIdf1BfDZwu6SAlhkg6Oh1DaPVFSdXpv9Fs4Ja0/SfAZyXVpmMW/wE8GhFLSLredpf0ZUkDJQ2TdFBnxUraS9Lh6eO9QfL30d57ZQVwQFhPuQzYHngF+D3JgGZPqCcZ9F0OfIPkA+/NzjZKP7A/DhxFUvOVwEkRsShd5TPAkrS77PT0eQDGkQzYriEZZ7gyIu5r52keJnlPWvcWniT5oJwLIGl0ekTS6KyNJV0g6dclTf8J/GvaffTVjNfUBHyeZJD4VZKusBltVruJZGD+WZJuo2+k294DfA24jWTP7D3Aiemy1cBHSN6vF4FngA+385pLDQS+SfL+vkiyN3Z+GdtZD5FvGGR9iaRbgEURkfseTKWRT7SzNrwHYds0SZMkvUdSP0lTSQ4Tvb3ouswqgc+ktm3dbsDPSAZUW4AvRMQfiy3JrDK4i8nMzDK5i8nMzDJtM11Mu+yyS9TU1BRdhplZRZk3b94rETEia9k2ExA1NTU0NTUVXYaZWUWR1NzeMncxmZlZJgeEmZllckCYmVmmbWYMwsx6r3Xr1tHS0sIbb7zR+cqWi0GDBlFdXc2AAQPK3sYBYWa5a2lpYdiwYdTU1ND+fZ8sLxHB8uXLaWlpYezYsWVv1+e7mBoboaYG+vVLfjc2draFmXXVG2+8wfDhwx0OBZHE8OHDu7wH16f3IBobYeZMWJveTqa5OZkHqK9vfzsz6zqHQ7G25P3v03sQs2e/HQ6t1q5N2s3M+ro+HRBLl3at3cwq0/Lly6mtraW2tpbddtuNkSNHbpp/6623Oty2qamJM888s9PnOPTQQ7ul1vvvv59jjjmmWx5ra/XpgBideRuW9tvNrGd099jg8OHDWbBgAQsWLOD000/n7LPP3jS/3XbbsX79+na3raur4/LLL+/0OR5++OGtK7IX6tMBccklMHjwO9sGD07azawYrWODzc0Q8fbYYHcfQDJjxgxOP/10DjroIM4991z+8Ic/cMghhzBx4kQOPfRQnn76aeCd3+gvuugiTjnlFA477DD23HPPdwTH0KFDN61/2GGHcfzxx7P33ntTX19P61Wz77jjDvbee28OOOAAzjzzzE73FFasWMFxxx3HhAkTOPjgg3n88ccBeOCBBzbtAU2cOJHVq1fzwgsvMGXKFGpra9l333158MEHt/o96tOD1K0D0bNnJ91Ko0cn4eABarPidDQ22N3/N1taWnj44Yepqqpi1apVPPjgg/Tv35977rmHCy64gNtuu22zbRYtWsR9993H6tWr2WuvvfjCF76w2bkFf/zjH1m4cCF77LEHkydP5ne/+x11dXWcdtppzJ07l7FjxzJ9+vRO67vwwguZOHEit99+O/feey8nnXQSCxYs4NJLL+WKK65g8uTJrFmzhkGDBtHQ0MCRRx7J7Nmz2bBhA2vbvolboE8HBCR/cA4Es96jJ8cGTzjhBKqqqgBYuXIlJ598Ms888wySWLduXeY2Rx99NAMHDmTgwIG8+93v5qWXXqK6uvod6xx44IGb2mpra1myZAlDhw5lzz333HQewvTp02loaOiwvoceemhTSB1++OEsX76cVatWMXnyZL7yla9QX1/PJz7xCaqrq5k0aRKnnHIK69at47jjjqO2tnar3hvo411MZtb79OTY4JAhQzZNf+1rX+PDH/4wTzzxBL/4xS/aPWdg4MCBm6arqqoyxy/KWWdrzJo1ix/+8Ie8/vrrTJ48mUWLFjFlyhTmzp3LyJEjmTFjBjfeeONWP48Dwsx6laLGBleuXMnIkSMBuP7667v98ffaay+effZZlixZAsAtt9zS6TYf+tCHaEwHX+6//3522WUXdthhB/7yl7+w3377cd555zFp0iQWLVpEc3Mzu+66K5///Oc59dRTmT9//lbX7IAws16lvh4aGmDMGJCS3w0N+XcFn3vuuZx//vlMnDix27/xA2y//fZceeWVTJ06lQMOOIBhw4ax4447drjNRRddxLx585gwYQKzZs3ihhtuAOCyyy5j3333ZcKECQwYMICjjjqK+++/n/3335+JEydyyy23cNZZZ211zdvMPanr6urCNwwy652eeuop3v/+9xddRuHWrFnD0KFDiQi++MUvMm7cOM4+++wee/6sfwdJ8yKiLmt970GYmfWQq6++mtraWvbZZx9WrlzJaaedVnRJHerzRzGZmfWUs88+u0f3GLaW9yDMrEdsK93ZlWpL3n8HhJnlbtCgQSxfvtwhUZDW+0EMGjSoS9u5i8nMclddXU1LSwvLli0rupQ+q/WOcl3hgDCz3A0YMKBLdzKz3sFdTGZmlskBYWZmmRwQZmaWKdeAkDRV0tOSFkualbF8iqT5ktZLOj5j+Q6SWiT9IM86zcxsc7kFhKQq4ArgKGA8MF3S+DarLQVmADe18zD/DszNq0YzM2tfnnsQBwKLI+LZiHgLuBmYVrpCRCyJiMeBjW03lnQAsCtwV441mplZO/IMiJHAcyXzLWlbpyT1A74LfLWT9WZKapLU5OOrzcy6V28dpP5n4I6IaOlopYhoiIi6iKgbMWJED5VmZtY35Hmi3PPAqJL56rStHIcAH5L0z8BQYDtJayJis4FuMzPLR54B8RgwTtJYkmA4EfincjaMiE23BpE0A6hzOJiZ9azcupgiYj1wBnAn8BTw04hYKOliSccCSJokqQU4AbhK0sK86jEzs67xHeXMzPow31HOzMy6zAFhZmaZHBBmZpbJAWFmZpkcEGZmlskBYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZpkcEGZmlskBYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZpkcEGZmlskBYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZpkcEGZmlskBYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZpkcEGZmlskBYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZplyDQhJUyU9LWmxpFkZy6dImi9pvaTjS9prJT0iaaGkxyV9Os86zcxsc7kFhKQq4ArgKGA8MF3S+DarLQVmADe1aV8LnBQR+wBTgcsk7ZRXrWZmtrn+OT72gcDiiHgWQNLNwDTgydYVImJJumxj6YYR8T8l03+T9DIwAngtx3rNzKxEnl1MI4HnSuZb0rYukXQgsB3wl4xlMyU1SWpatmzZFhdqZmab69WD1JJ2B34EfDYiNrZdHhENEVEXEXUjRozo+QLNzLZheQbE88CokvnqtK0sknYAfgXMjojfd3NtZmbWiTwD4jFgnKSxkrYDTgTmlLNhuv7PgRsj4tYcazQzs3bkFhARsR44A7gTeAr4aUQslHSxpGMBJE2S1AKcAFwlaWG6+aeAKcAMSQvSn9q8ajUzs80pIoquoVvU1dVFU1NT0WWYmVUUSfMioi5rWa8epDYzs+I4IMzMLJMDwszMMjkgzMwskwPCzMwyOSDMzCyTA8LMzDI5IMzMLJMDwszMMjkgzMwskwPCzMwyOSDMzCyTA8LMzDI5IMzMLJMDwszMMjkgzMwskwPCzMwyOSDMzCyTA8LMzDI5IMzMLJMDwszMMpUVEJKGSOqXTr9P0rGSBuRbmpmZFancPYi5wCBJI4G7gM8A1+dVlJmZFa/cgFBErAU+AVwZEScA++RXlpmZFa3sgJB0CFAP/Cptq8qnJDMz6w3KDYgvA+cDP4+IhZL2BO7LrywzMyta/3JWiogHgAcA0sHqVyLizDwLMzOzYpV7FNNNknaQNAR4AnhS0r/kW5qZmRWp3C6m8RGxCjgO+DUwluRIJjMz20aVGxAD0vMejgPmRMQ6IPIry8zMilZuQFwFLAGGAHMljQFW5VWUmZkVr9xB6suBy0uamiV9OJ+SzMysNyh3kHpHSd+T1JT+fJdkb8LMzLZR5XYxXQusBj6V/qwCrutsI0lTJT0tabGkWRnLp0iaL2m9pOPbLDtZ0jPpz8ll1tkrNDZCTQ3065f8bmwsuiIzs64rq4sJeE9EfLJk/uuSFnS0gaQq4ArgI0AL8JikORHxZMlqS4EZwFfbbPsu4EKgjmQwfF667atl1luYxkaYORPWrk3mm5uTeYD6+uLqMjPrqnL3IF6X9MHWGUmTgdc72eZAYHFEPBsRbwE3A9NKV4iIJRHxOLCxzbZHAndHxIo0FO4GppZZa6Fmz347HFqtXZu0m5lVknL3IE4HbpS0Yzr/KtBZt89I4LmS+RbgoDKfL2vbkW1XkjQTmAkwevToMh86X0uXdq3dzKy3KmsPIiL+FBH7AxOACRExETg818rKEBENEVEXEXUjRowouhwA2supXpJfZmZl69Id5SJiVXpGNcBXOln9eWBUyXx12laOrdm2UJdcAoMHv7Nt8OCk3cyskmzNLUfVyfLHgHGSxkraDjgRmFPmY98JfFTSzpJ2Bj6atvV69fXQ0ABjxoCU/G5o8AC1mVWecscgsnR4qY2IWC/pDJIP9irg2vRS4RcDTRExR9Ik4OfAzsDHJX09IvaJiBWS/p0kZAAujogVW1Frj6qvdyCYWeVTRPuf85JWkx0EAraPiK0JmG5VV1cXTU1NRZdhZlZRJM2LiLqsZR1+wEfEsHxKMjOz3m5rxiDMzGwb5oAwM7NMDggzM8vkgDAzs0wOCDMzy+SAMDOzTA4IMzPL5IAwM7NMDggzM8vkgDAzs0wOCDMzy+SAqCCNjVBTA/36Jb8bG4uuyMy2Zb3maqzWscZGmDnz7ftdNzcn8+BLi5tZPrwHUSFmz347HFqtXZu0m5nlwQFRIZYu7Vq7mdnWckBUiNGju9ZuZra1HBAV4pJLYPDgd7YNHpy0m5nlwQFRIerroaEBxowBKfnd0OABajPLj49iqiD19Q4EM+s53oMwM7NMDggzM8vkgDAzs0wOCDMzy+SAMF/jycwy+SimPs7XeDKz9ngPoo/zNZ7MrD0OiD7O13gys/Y4IPo4X+PJzNrjgOjjfI0nM2uPA6KP8zWezKw9PorJfI0nM8vkPQgzM8vkgDAzs0y5BoSkqZKelrRY0qyM5QMl3ZIuf1RSTdo+QNINkv4s6SlJ5+dZp+XDZ2ibVbbcAkJSFXAFcBQwHpguaXyb1T4HvBoR7wW+D3wrbT8BGBgR+wEHAKe1hodVhtYztJubIeLtM7QdEmaVI889iAOBxRHxbES8BdwMTGuzzjTghnT6VuAISQICGCKpP7A98BawKsdarZv5DG2zypdnQIwEniuZb0nbMteJiPXASmA4SVj8HXgBWApcGhEr2j6BpJmSmiQ1LVu2rPtfgW0xn6FtVvl66yD1gcAGYA9gLHCOpD3brhQRDRFRFxF1I0aM6OkarQM+Q9us8uUZEM8Do0rmq9O2zHXS7qQdgeXAPwG/iYh1EfEy8DugLsdarZv5DG2zypdnQDwGjJM0VtJ2wInAnDbrzAFOTqePB+6NiCDpVjocQNIQ4GBgUY61WjfzGdpmlS+3M6kjYr2kM4A7gSrg2ohYKOlioCki5gDXAD+StBhYQRIikBz9dJ2khYCA6yLi8bxqtXz4DG2zyqbkC3vlq6uri6ampqLLsJw1NiZHQi1dmoxnXHKJQ8hsa0iaFxGZXfi+FpNVDN/9zqxn9dajmMw243MrzHqWA8Iqhs+tMOtZDgirGD63wqxnOSCsYvjcCrOe5YCwiuFzK8x6lgPCKkp9PSxZAhs3Jr+7Kxx8aXKzzfkwV+vzfPisWTbvQVif58NnzbI5IKzP8+GzZtkcENbn+fBZs2wOCOvzfPisWTYHhPV5eR4+66OjrJL5KCYz8rk0uY+OskrnPQiznPjoKKt0DgiznPjoKKt0DgiznPjoKKt0DgiznOR5dJQHv60nOCDMcpLX0VGtg9/NzRDx9uC3Q8K6m+9JbVZhamqSUGhrzJjkAoZmXdHRPam9B2FWYTz4bT3FAWFWYTz4bT3FAWFWYfIa/PbAt7XlgDCrMHkMfnvg27J4kNrMPPDdh3mQ2sw65IFvy+KAMLNcB749tlG5HBBmluvAt8c2KpcDwsxyO+vbV7StbB6kNrPc9OuX7Dm0JcHGjT1fj23Og9RmVgiPbVQ2B4SZ5cZjG5XNAWFmufHYRmXzGISZVRyPbXSfwsYgJE2V9LSkxZJmZSwfKOmWdPmjkmpKlk2Q9IikhZL+LGlQnrWaWeXw2EbPyC0gJFUBVwBHAeOB6ZLGt1ntc8CrEfFe4PvAt9Jt+wM/Bk6PiH2Aw4B1edVqZpXFYxs9I889iAOBxRHxbES8BdwMTGuzzjTghnT6VuAISQI+CjweEX8CiIjlEbEhx1rNrIJ4bKNn5BkQI4HnSuZb0rbMdSJiPbASGA68DwhJd0qaL+ncrCeQNFNSk6SmZcuWdfsLMLPeq74+uZDgxo3J760NB8jvmlSV2m3VW49i6g98EKhPf/+jpCParhQRDRFRFxF1I0aM6OkazWwbk8fYRiV3W+UZEM8Do0rmq9O2zHXScYcdgeUkextzI+KViFgL3AF8IMdazcxyGduo5G6rPAPiMWCcpLGStgNOBOa0WWcOcHI6fTxwbyTH3d4J7CdpcBoc/wA8mWOtZma5jG3keSn1vLuu+nfvw70tItZLOoPkw74KuDYiFkq6GGiKiDnANcCPJC0GVpCECBHxqqTvkYRMAHdExK/yqtXMrFV9ffeMZ7QaPTr7Zkxbe0hua9dV695Ja9cVdF/9PlHOzCxHbT/IIem22to9k+66C6Av1mdmVpC8DsntibsA5tbFZGZmie7utoL8uq5KeQ/CzKwC5XU2eSkHhJlZBcqr66qUu5jMzCpUHl1XpbwHYWZmmRwQZmaWyQFhZmaZHBBmZpbJAWFmZpm2mUttSFoGZJw2UqhdgFeKLqILKqneSqoVKqveSqoVKqve3ljrmIjIvF/CNhMQvZGkpvaucdIbVVK9lVQrVFa9lVQrVFa9lVQruIvJzMza4YAwM7NMDoh8NRRdQBdVUr2VVCtUVr2VVCtUVr2VVKvHIMzMLJv3IMzMLJMDwszMMjkgciBplKT7JD0paaGks4quqTOSqiT9UdIvi66lM5J2knSrpEWSnpJ0SNE1tUfS2enfwBOSfiJpUNE1lZJ0raSXJT1R0vYuSXdLeib9vXORNZZqp97vpH8Lj0v6uaSdiqyxVVatJcvOkRSSdimitnI5IPKxHjgnIsYDBwNflDS+4Jo6cxbwVNFFlOn/AL+JiL2B/emldUsaCZwJ1EXEvkAVcGKxVW3memBqm7ZZwG8jYhzw23S+t7iezeu9G9g3IiYA/wOc39NFteN6Nq8VSaOAjwLdeHPQfDggchARL0TE/HR6NckH2Mhiq2qfpGrgaOCHRdfSGUk7AlOAawAi4q2IeK3YqjrUH9heUn9gMPC3gut5h4iYC6xo0zwNuCGdvgE4rkeL6kBWvRFxV0SsT2d/D1T3eGEZ2nlvAb4PnAv0+iOEHBA5k1QDTAQeLbaSDl1G8ge7sehCyjAWWAZcl3aJ/VDSkKKLyhIRzwOXknxTfAFYGRF3FVtVWXaNiBfS6ReBXYsspotOAX5ddBHtkTQNeD4i/lR0LeVwQORI0lDgNuDLEbGq6HqySDoGeDki5hVdS5n6Ax8A/m9ETAT+Tu/qAtkk7bufRhJqewBDJP3vYqvqmkiOg+/133QBJM0m6d5tLLqWLJIGAxcA/1Z0LeVyQORE0gCScGiMiJ8VXU8HJgPHSloC3AwcLunHxZbUoRagJSJa98huJQmM3uh/AX+NiGURsQ74GXBowTWV4yVJuwOkv18uuJ5OSZoBHAPUR+89ues9JF8W/pT+f6sG5kvardCqOuCAyIEkkfSRPxUR3yu6no5ExPkRUR0RNSQDqPdGRK/9lhsRLwLPSdorbToCeLLAkjqyFDhY0uD0b+IIeumAehtzgJPT6ZOB/1dgLZ2SNJWki/TYiFhbdD3tiYg/R8S7I6Im/f/WAnwg/ZvulRwQ+ZgMfIbk2/iC9OdjRRe1DfkS0CjpcaAW+I+C68mU7uXcCswH/kzy/61XXWpB0k+AR4C9JLVI+hzwTeAjkp4h2Qv6ZpE1lmqn3h8Aw4C70/9r/11okal2aq0ovtSGmZll8h6EmZllckCYmVkmB4SZmWVyQJiZWSYHhJmZZXJAmHVC0oaSw5UXSOq2M7cl1WRd7dOsN+hfdAFmFeD1iKgtugiznuY9CLMtJGmJpG9L+rOkP0h6b9peI+ne9P4Ev5U0Om3fNb1fwZ/Sn9bLblRJujq9b8RdkrZP1z8zvafI45JuLuhlWh/mgDDr3PZtupg+XbJsZUTsR3I272Vp238BN6T3J2gELk/bLwceiIj9Sa4ftTBtHwdcERH7AK8Bn0zbZwET08c5Pa8XZ9Yen0lt1glJayJiaEb7EuDwiHg2vTjjixExXNIrwO4RsS5tfyEidpG0DKiOiDdLHqMGuDu9OQ+SzgMGRMQ3JP0GWAPcDtweEWtyfqlm7+A9CLOtE+1Md8WbJdMbeHts8GjgCpK9jcfSmw6Z9RgHhNnW+XTJ70fS6Yd5+9ai9cCD6fRvgS/ApnuA79jeg0rqB4yKiPuA84Adgc32Yszy5G8kZp3bXtKCkvnfRETroa47p1eVfROYnrZ9ieSOd/9Ccve7z6btZwEN6VU9N5CExQtkqwJ+nIaIgMt7+a1VbRvkMQizLZSOQdRFxCtF12KWB3cxmZlZJu9BmJlZJu9BmJlZJgeEmZllckCYmVkmB4SZmWVyQJiZWab/DwFqX62UA1NLAAAAAElFTkSuQmCC\n"},"metadata":{"needs_background":"light"}},{"output_type":"display_data","data":{"text/plain":[""],"image/png":"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\n"},"metadata":{"needs_background":"light"}}]},{"cell_type":"markdown","source":["# Tensori"],"metadata":{"id":"DCcTH7De7Dva"}},{"cell_type":"code","source":["# Scalari (tensori a zero dimensioni)\n","\n","x = np.array(12)"],"metadata":{"id":"EsAWa_NF7KD4"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["x"],"metadata":{"id":"rEaq-bCy7SNv","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948501014,"user_tz":-60,"elapsed":6,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"d5bc1076-8f2f-475c-b5e7-fd0a0f3e9a6c"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array(12)"]},"metadata":{},"execution_count":40}]},{"cell_type":"code","source":["x.shape"],"metadata":{"id":"iW4A2Mid7T0t","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948502835,"user_tz":-60,"elapsed":254,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"01ac9b80-7039-4991-ee7f-b24feb25d4f9"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["()"]},"metadata":{},"execution_count":41}]},{"cell_type":"code","source":["# Vettori (tensori ad una dimensione)\n","\n","x = np.array([12, 3, 6, 14])"],"metadata":{"id":"Q_djOScz7VIM"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["x"],"metadata":{"id":"Jw04eg6s7ed4","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948506107,"user_tz":-60,"elapsed":243,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"905803cd-085a-4828-e6b5-4369ef5b3f86"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([12, 3, 6, 14])"]},"metadata":{},"execution_count":43}]},{"cell_type":"code","source":["x.shape"],"metadata":{"id":"FRVuy9Wy7fFd","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948508433,"user_tz":-60,"elapsed":12,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"618504e3-2798-4740-99f8-37f07e1ea4ee"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(4,)"]},"metadata":{},"execution_count":44}]},{"cell_type":"code","source":["# Matrici (tensori a due dimensioni)\n","\n","x = np.array([[5, 78, 2, 34, 0],\n"," [6, 79, 3, 35, 1],\n"," [7, 80, 4, 36, 2]])"],"metadata":{"id":"m7krI7_X7f2N"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["x"],"metadata":{"id":"iLwrkM6O7nnu","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948511367,"user_tz":-60,"elapsed":349,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"155baab0-e9f7-4dd9-ccc7-462b791b6f4e"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[ 5, 78, 2, 34, 0],\n"," [ 6, 79, 3, 35, 1],\n"," [ 7, 80, 4, 36, 2]])"]},"metadata":{},"execution_count":46}]},{"cell_type":"code","source":["x.shape"],"metadata":{"id":"0Khr7rGk7n7d","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948513613,"user_tz":-60,"elapsed":262,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"73995714-442b-4753-9e58-244b668b3eb1"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(3, 5)"]},"metadata":{},"execution_count":47}]},{"cell_type":"code","source":["# Tensori a tre dimensioni\n","\n","x = np.array(\n"," [\n"," [\n"," [ 5, 78, 2, 34, 0],\n"," [ 6, 79, 3, 35, 1],\n"," [ 7, 80, 4, 36, 2]\n"," ],\n"," [\n"," [ 1, 0, 0, 1, 0],\n"," [ 0, 1, 1, 0, 1],\n"," [ 1, 0, 0, 0, 0]\n"," ] \n","])"],"metadata":{"id":"OndeKcrr7ot3"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["x"],"metadata":{"id":"BFamLGLZ8RP4","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948516875,"user_tz":-60,"elapsed":4,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"f9150a50-9b0d-47c7-e579-79e6c2d1ad81"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[[ 5, 78, 2, 34, 0],\n"," [ 6, 79, 3, 35, 1],\n"," [ 7, 80, 4, 36, 2]],\n","\n"," [[ 1, 0, 0, 1, 0],\n"," [ 0, 1, 1, 0, 1],\n"," [ 1, 0, 0, 0, 0]]])"]},"metadata":{},"execution_count":49}]},{"cell_type":"code","source":["x.shape"],"metadata":{"id":"Tc2Km2bE8SKl","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1647948518025,"user_tz":-60,"elapsed":4,"user":{"displayName":"Gianluca Amato","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GhEmlMukD8qDnxG1PlDxK7Dzm1rjUKZSz0BhiJc9w=s64","userId":"18269286707108730791"}},"outputId":"b59e9968-44f0-4ab1-c4d0-34478c8e2fde"},"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(2, 3, 5)"]},"metadata":{},"execution_count":50}]}]}