{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"## Import the libraries\n","metadata":{"execution":{"iopub.status.busy":"2022-02-11T15:15:42.09471Z","iopub.execute_input":"2022-02-11T15:15:42.095103Z","iopub.status.idle":"2022-02-11T15:15:42.099724Z","shell.execute_reply.started":"2022-02-11T15:15:42.095016Z","shell.execute_reply":"2022-02-11T15:15:42.098645Z"}}},{"cell_type":"code","source":"!pip install pyunpack\n!pip install patool\n!pip install py7zr\n!pip install sounddevice\n!pip install noisereduce\n!pip install librosa\n! pip install python_speech_features\n! pip install tensorflow==2.4\n! pip install malaya_speech\n! pip install webrtcvad","metadata":{"execution":{"iopub.status.busy":"2022-06-25T20:58:14.208024Z","iopub.execute_input":"2022-06-25T20:58:14.208438Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pip install py7zr","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#pip update huggingface_hub\n! pip install --upgrade transformers","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **importing libraries**","metadata":{}},{"cell_type":"code","source":"pip install git+https://github.com/huggingface/transformers","metadata":{"execution":{"iopub.status.idle":"2022-06-25T21:01:29.626273Z","shell.execute_reply.started":"2022-06-25T21:00:47.883237Z","shell.execute_reply":"2022-06-25T21:01:29.625285Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np                        # linear algebra library\nimport pandas as pd                       # data frames processing\nimport matplotlib.pyplot as plt          # visualization library\nimport seaborn as sn                      # visualization library\n\n\n# audio processing library\nimport librosa                          \nimport IPython.display as ipd            \nfrom scipy.io import wavfile\nimport noisereduce as nr\nfrom malaya_speech import Pipeline\nimport malaya_speech\nfrom python_speech_features import mfcc\nfrom sklearn.preprocessing import LabelEncoder\n\n\n# unpacking the dataset\nfrom py7zr import unpack_7zarchive    \n\n#operating system libraries\nimport shutil\nimport warnings\nwarnings.filterwarnings(\"ignore\")\nimport tensorflow \nimport os","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:01:29.627897Z","iopub.execute_input":"2022-06-25T21:01:29.628233Z","iopub.status.idle":"2022-06-25T21:01:34.640095Z","shell.execute_reply.started":"2022-06-25T21:01:29.628202Z","shell.execute_reply":"2022-06-25T21:01:34.639177Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# unpacking the dataset\nshutil.register_unpack_format('7zip', ['.7z'], unpack_7zarchive)\nshutil.unpack_archive('/kaggle/input/tensorflow-speech-recognition-challenge/train.7z', '/kaggle/working/tensorflow-speech-recognition-challenge/train/')","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:01:34.641313Z","iopub.execute_input":"2022-06-25T21:01:34.641611Z","iopub.status.idle":"2022-06-25T21:10:56.758220Z","shell.execute_reply.started":"2022-06-25T21:01:34.641585Z","shell.execute_reply":"2022-06-25T21:10:56.757337Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# <center> Implementing the Speech Recognition Model in Python\n**Dataset used for our Speech Recognition Project**\n    \nIt is a set of 10 numbers each is repeated 2000 times with different accents and different back ground conditions. TensorFlow recently released the Speech Commands Datasets. It includes 65,000 one-second long utterances of 30 short words, by thousands of different people. We’ll build a speech recognition system that understands simple spoken commands. <br>    \n    \n__You can download the dataset from__ [here](https://www.kaggle.com/c/tensorflow-speech-recognition-challenge).\n","metadata":{}},{"cell_type":"markdown","source":"**Data Exploration and Visualization**\n\nData Exploration and Visualization helps us to understand the data as well as pre-processing steps in a better way. \n\n","metadata":{}},{"cell_type":"code","source":"train_audio_path = '/kaggle/working/tensorflow-speech-recognition-challenge/train/train/audio/' #path of the training data","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:56.760675Z","iopub.execute_input":"2022-06-25T21:10:56.761169Z","iopub.status.idle":"2022-06-25T21:10:56.765329Z","shell.execute_reply.started":"2022-06-25T21:10:56.761132Z","shell.execute_reply":"2022-06-25T21:10:56.764501Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Accessing a sample file in data**","metadata":{}},{"cell_type":"code","source":"samples, sample_rate = librosa.load(train_audio_path+'on/5a3712c9_nohash_1.wav', sr = 16000)  # loading a sample to be explored carefully\nfig = plt.figure(figsize=(14, 8))\nax1 = fig.add_subplot(211)\nax1.set_title('Raw wave of ' + '../input/train/audio/on/0a7c2a8d_nohash_0.wav')\nax1.set_xlabel('time')\nax1.set_ylabel('Amplitude')\nax1.plot(np.linspace(0, sample_rate/len(samples), sample_rate), samples)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:56.767967Z","iopub.execute_input":"2022-06-25T21:10:56.768291Z","iopub.status.idle":"2022-06-25T21:10:56.985814Z","shell.execute_reply.started":"2022-06-25T21:10:56.768258Z","shell.execute_reply":"2022-06-25T21:10:56.984975Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fs=16000\nipd.Audio(samples, rate=fs)  #listen to audio file before noise reduction\nprint(\"Sample Rate:\",fs)\nsr=fs","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:56.987611Z","iopub.execute_input":"2022-06-25T21:10:56.988106Z","iopub.status.idle":"2022-06-25T21:10:56.994879Z","shell.execute_reply.started":"2022-06-25T21:10:56.988067Z","shell.execute_reply":"2022-06-25T21:10:56.993909Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Noise Reduction\ntime = np.linspace(0, len(samples - 1) / fs, len(samples - 1))\nreduced_noise1 = nr.reduce_noise(y=samples, sr=fs,stationary=True)\nplt.plot(time, reduced_noise1)  # plot in seconds\nplt.xlabel(\"Time [seconds]\")\nplt.ylabel(\"Voice amplitude\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:56.996151Z","iopub.execute_input":"2022-06-25T21:10:56.996673Z","iopub.status.idle":"2022-06-25T21:10:57.734439Z","shell.execute_reply.started":"2022-06-25T21:10:56.996636Z","shell.execute_reply":"2022-06-25T21:10:57.733581Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ipd.Audio(reduced_noise1, rate=sample_rate)  #listen to audio file after noise reduction","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:57.735714Z","iopub.execute_input":"2022-06-25T21:10:57.736192Z","iopub.status.idle":"2022-06-25T21:10:57.747206Z","shell.execute_reply.started":"2022-06-25T21:10:57.736154Z","shell.execute_reply":"2022-06-25T21:10:57.746380Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#Silence Removal\nvad = malaya_speech.vad.webrtc()\ny=reduced_noise1\ny_= malaya_speech.resample(y, sr, 16000)\ny_ = malaya_speech.astype.float_to_int(y_)\nframes = malaya_speech.generator.frames(y, 30, sr)\nframes_ = list(malaya_speech.generator.frames(y_, 30, 16000, append_ending_trail = False))\nframes_webrtc = [(frames[no], vad(frame)) for no, frame in enumerate(frames_)]\ny_ = malaya_speech.combine.without_silent(frames_webrtc)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:57.748593Z","iopub.execute_input":"2022-06-25T21:10:57.749270Z","iopub.status.idle":"2022-06-25T21:10:57.911081Z","shell.execute_reply.started":"2022-06-25T21:10:57.749233Z","shell.execute_reply":"2022-06-25T21:10:57.910252Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ipd.Audio(y_, rate = sr )     #listen to audio file after noise reduction","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:57.912244Z","iopub.execute_input":"2022-06-25T21:10:57.912586Z","iopub.status.idle":"2022-06-25T21:10:57.919445Z","shell.execute_reply.started":"2022-06-25T21:10:57.912552Z","shell.execute_reply":"2022-06-25T21:10:57.918687Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# padding signal with zeros\nzero = np.zeros((1*sr-y_.shape[0]))\nsignal = np.concatenate((y_,zero))\nsignal.shape\ntime = np.linspace(0, len(signal - 1) / fs, len(signal - 1))","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:57.920715Z","iopub.execute_input":"2022-06-25T21:10:57.921236Z","iopub.status.idle":"2022-06-25T21:10:57.931461Z","shell.execute_reply.started":"2022-06-25T21:10:57.921199Z","shell.execute_reply":"2022-06-25T21:10:57.930693Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Visualization of Audio signal in time series domain**\n\nNow, we’ll visualize the audio signal in the time series domain:","metadata":{}},{"cell_type":"code","source":"plt.plot(time,signal)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:57.933167Z","iopub.execute_input":"2022-06-25T21:10:57.933404Z","iopub.status.idle":"2022-06-25T21:10:58.079929Z","shell.execute_reply.started":"2022-06-25T21:10:57.933381Z","shell.execute_reply":"2022-06-25T21:10:58.078998Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"labels=os.listdir(train_audio_path)     #Extracting labels to determine classes","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:58.081315Z","iopub.execute_input":"2022-06-25T21:10:58.081687Z","iopub.status.idle":"2022-06-25T21:10:58.086218Z","shell.execute_reply.started":"2022-06-25T21:10:58.081650Z","shell.execute_reply":"2022-06-25T21:10:58.085032Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Distribution of the Data set**","metadata":{}},{"cell_type":"code","source":"#find count of each label and plot bar graph\nno_of_recordings=[]\nfor label in labels:\n    waves = [f for f in os.listdir(train_audio_path + '/'+ label) if f.endswith('.wav')]\n    no_of_recordings.append(len(waves))\n    \n#plot\nplt.figure(figsize=(30,5))\nindex = np.arange(len(labels))\nplt.bar(index, no_of_recordings)\nplt.xlabel('Commands', fontsize=12)\nplt.ylabel('No of recordings', fontsize=12)\nplt.xticks(index, labels, fontsize=15, rotation=60)\nplt.title('No. of recordings for each command')\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:58.087673Z","iopub.execute_input":"2022-06-25T21:10:58.088039Z","iopub.status.idle":"2022-06-25T21:10:58.424205Z","shell.execute_reply.started":"2022-06-25T21:10:58.088004Z","shell.execute_reply":"2022-06-25T21:10:58.423369Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Words used as Classes for data set","metadata":{}},{"cell_type":"code","source":"labels=[\"zero\",\"one\",\"two\",\"three\",\"four\",\"five\",\"six\",\"seven\",\"eight\",\"nine\"]","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:58.425547Z","iopub.execute_input":"2022-06-25T21:10:58.426076Z","iopub.status.idle":"2022-06-25T21:10:58.430758Z","shell.execute_reply.started":"2022-06-25T21:10:58.426037Z","shell.execute_reply":"2022-06-25T21:10:58.429798Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Preprocessing the audio waves**\n\nlet us read the audio waves and use the below-preprocessing steps :\n\n* Noise Reduction\n* Silence Removal\n* Extracting MFCCs\n","metadata":{}},{"cell_type":"code","source":"sr=16000    # sample rate\nvad = malaya_speech.vad.webrtc()\nall_wave = []     #intitialize array to stack wave files of the whole data set in it \nall_label = []    #intitialize array to stack label of wave files of the whole data set in it \nfor label in labels:\n    print(label)\n    waves = [f for f in os.listdir(train_audio_path + '/'+ label) if f.endswith('.wav')] # access on each file\n    for wav in waves:\n        samples, sample_rate = librosa.load(train_audio_path + '/' + label + '/' + wav, sr = 16000)\n        samples = nr.reduce_noise(y=samples, sr=sr,stationary=True)  #noise reduction\n        y_= malaya_speech.resample(samples, sr, 16000)               # silence removal\n        y_ = malaya_speech.astype.float_to_int(y_)\n        frames = malaya_speech.generator.frames(samples, 30, sr)\n        frames_ = list(malaya_speech.generator.frames(y_, 30, 16000, append_ending_trail = False))\n        frames_webrtc = [(frames[no], vad(frame)) for no, frame in enumerate(frames_)]\n        y_ = malaya_speech.combine.without_silent(frames_webrtc)\n        zero = np.zeros(((1*sr+4000)-y_.shape[0]))                 \n        signal = np.concatenate((y_,zero))     # concatenation with zeros to adust length of the vector\n        all_wave.append(signal)     #append waves one by one \n        all_label.append(label)     #append corresponding label one by one","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:10:58.432364Z","iopub.execute_input":"2022-06-25T21:10:58.432896Z","iopub.status.idle":"2022-06-25T21:22:47.048423Z","shell.execute_reply.started":"2022-06-25T21:10:58.432859Z","shell.execute_reply":"2022-06-25T21:22:47.047560Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"shape of waves array\",np.array(all_wave).shape)\nprint(\"shape of labels array\",np.array(all_label).shape)\n\n#Inspecting random sample\ntime = np.linspace(0, len(signal - 1) / fs, len(signal - 1))\nplt.plot(time,np.array(all_wave)[2000,:])\nprint(np.array(all_label)[2000])\nipd.Audio(np.array(all_wave)[2000,:], rate = sr )","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:22:47.049794Z","iopub.execute_input":"2022-06-25T21:22:47.050313Z","iopub.status.idle":"2022-06-25T21:22:57.603592Z","shell.execute_reply.started":"2022-06-25T21:22:47.050275Z","shell.execute_reply":"2022-06-25T21:22:57.602732Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Extracting MFCCs","metadata":{}},{"cell_type":"code","source":"all_mfcc=[]    #intitialize array to stack MFCCs of the whole data set in it\nfor wave in all_wave:\n    i=0\n    mfcc_feat = mfcc(wave , fs, winlen=256/fs, winstep=256/(2*fs), numcep=13, nfilt=26, nfft=256,\n                 lowfreq=0, highfreq=fs/2, preemph=0.97, ceplifter=22, appendEnergy=True, winfunc=np.hamming)\n    mfcc_feat= np.transpose(mfcc_feat)\n    all_mfcc.append(mfcc_feat)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:22:57.604887Z","iopub.execute_input":"2022-06-25T21:22:57.605410Z","iopub.status.idle":"2022-06-25T21:24:09.199883Z","shell.execute_reply.started":"2022-06-25T21:22:57.605368Z","shell.execute_reply":"2022-06-25T21:24:09.198768Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"shape of MFCCs\",np.array(all_mfcc).shape)\nprint(\"shape of Corresponding lables\",np.array(all_label).shape)\n\n# dimensions of the data\nd1=np.array(all_mfcc).shape[1]\nd2=np.array(all_mfcc).shape[2]\nd=d1*d2","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:24:09.201579Z","iopub.execute_input":"2022-06-25T21:24:09.202212Z","iopub.status.idle":"2022-06-25T21:24:09.726552Z","shell.execute_reply.started":"2022-06-25T21:24:09.202171Z","shell.execute_reply":"2022-06-25T21:24:09.725698Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"op_mfcc=np.array(all_mfcc)     # transform list to array to be fed to the model\nop_mfcc=op_mfcc.reshape(np.array(all_mfcc).shape[0],-1)   # adjust shape of the samples\nop_mfcc.shape","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:24:09.730531Z","iopub.execute_input":"2022-06-25T21:24:09.730796Z","iopub.status.idle":"2022-06-25T21:24:10.061525Z","shell.execute_reply.started":"2022-06-25T21:24:09.730770Z","shell.execute_reply":"2022-06-25T21:24:10.060603Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## One hot encoding","metadata":{}},{"cell_type":"code","source":"#all_label = all_label.tolist()\nle = LabelEncoder()\ny=le.fit_transform(all_label)\nclasses= list(le.classes_)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:24:10.063511Z","iopub.execute_input":"2022-06-25T21:24:10.063867Z","iopub.status.idle":"2022-06-25T21:24:10.076907Z","shell.execute_reply.started":"2022-06-25T21:24:10.063830Z","shell.execute_reply":"2022-06-25T21:24:10.076097Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Model based on ANN** ","metadata":{}},{"cell_type":"code","source":"! pip install --upgrade tensorflow\n! pip install --upgrade tensorflow-gpu\n! pip install keras==2.3.1","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:24:10.078222Z","iopub.execute_input":"2022-06-25T21:24:10.078598Z","iopub.status.idle":"2022-06-25T21:26:21.454827Z","shell.execute_reply.started":"2022-06-25T21:24:10.078563Z","shell.execute_reply":"2022-06-25T21:26:21.453749Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Importing model libraries from keras\nfrom keras.optimizers import SGD\nfrom keras.constraints import maxnorm\nfrom tensorflow.keras import Sequential\nfrom tensorflow.keras.layers import Flatten, Dense,Dropout\nfrom keras.callbacks import EarlyStopping, ModelCheckpoint\nfrom sklearn.model_selection import train_test_split\n","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:21.457338Z","iopub.execute_input":"2022-06-25T21:26:21.457708Z","iopub.status.idle":"2022-06-25T21:26:21.567649Z","shell.execute_reply.started":"2022-06-25T21:26:21.457667Z","shell.execute_reply":"2022-06-25T21:26:21.566731Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y=tensorflow.keras.utils.to_categorical(y, num_classes=len(labels), dtype='float32')  # one hot encoded varibales to categeorical values\ny.shape","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:21.571438Z","iopub.execute_input":"2022-06-25T21:26:21.573524Z","iopub.status.idle":"2022-06-25T21:26:21.585078Z","shell.execute_reply.started":"2022-06-25T21:26:21.573486Z","shell.execute_reply":"2022-06-25T21:26:21.584375Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_tr, x_val, y_tr, y_val= train_test_split(op_mfcc,np.array(y),stratify=y,test_size = 0.2,random_state=777,shuffle=True)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:21.586312Z","iopub.execute_input":"2022-06-25T21:26:21.586869Z","iopub.status.idle":"2022-06-25T21:26:22.623072Z","shell.execute_reply.started":"2022-06-25T21:26:21.586834Z","shell.execute_reply":"2022-06-25T21:26:22.622198Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Inspect shapes of each set\nprint(\"shape of training samples\",x_tr.shape)\nprint(\"shape of training labels\",y_tr.shape)\nprint(\"shape of test samples\",x_val.shape)\nprint(\"shape of test labels\",y_val.shape)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:22.624527Z","iopub.execute_input":"2022-06-25T21:26:22.625104Z","iopub.status.idle":"2022-06-25T21:26:22.644549Z","shell.execute_reply.started":"2022-06-25T21:26:22.625065Z","shell.execute_reply":"2022-06-25T21:26:22.643622Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### **Model Architecture**","metadata":{}},{"cell_type":"code","source":"#Model Architecture\nmodel = Sequential()\nmodel.add(Dense(100, activation='relu', input_shape=(d,), kernel_constraint=maxnorm(3)))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(80, activation='relu', kernel_constraint=maxnorm(3)))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(len(classes), activation='softmax' , kernel_constraint=maxnorm(3)))","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:22.645799Z","iopub.execute_input":"2022-06-25T21:26:22.646684Z","iopub.status.idle":"2022-06-25T21:26:26.827083Z","shell.execute_reply.started":"2022-06-25T21:26:22.646651Z","shell.execute_reply":"2022-06-25T21:26:26.826267Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tensorflow.keras.utils.plot_model(model, 'model.png',show_shapes=True)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:26:26.828399Z","iopub.execute_input":"2022-06-25T21:26:26.828794Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.compile(loss='categorical_crossentropy',optimizer='adamax',metrics=['accuracy'])\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=10, min_delta=0.0001) \nmc = ModelCheckpoint('best_model.hdf5', monitor='val_acc', verbose=1, save_best_only=True, mode='max')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history=model.fit(x_tr, y_tr,validation_data=(x_val,y_val), epochs=300, batch_size=65)","metadata":{"execution":{"iopub.status.idle":"2022-06-25T21:31:23.879127Z","shell.execute_reply.started":"2022-06-25T21:26:27.621538Z","shell.execute_reply":"2022-06-25T21:31:23.878281Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Model Evaluation","metadata":{}},{"cell_type":"code","source":"train_score = model.evaluate(x_tr, y_tr, batch_size=12)\nprint(train_score)\n\nprint('----------------Training Complete-----------------')\n\ntest_score = model.evaluate(x_val, y_val, batch_size = 12)\nprint(test_score)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:23.880500Z","iopub.execute_input":"2022-06-25T21:31:23.880794Z","iopub.status.idle":"2022-06-25T21:31:27.614148Z","shell.execute_reply.started":"2022-06-25T21:31:23.880766Z","shell.execute_reply":"2022-06-25T21:31:27.613197Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history.history.keys()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:27.616358Z","iopub.execute_input":"2022-06-25T21:31:27.616936Z","iopub.status.idle":"2022-06-25T21:31:27.623086Z","shell.execute_reply.started":"2022-06-25T21:31:27.616896Z","shell.execute_reply":"2022-06-25T21:31:27.622288Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['loss'], label='train')  # losses learning curve of training set.\nplt.plot(history.history['val_loss'], label='test') # losses learning curve of validation set.\nplt.legend()\nplt.title(\"losses learning curves\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:27.624431Z","iopub.execute_input":"2022-06-25T21:31:27.624869Z","iopub.status.idle":"2022-06-25T21:31:27.770295Z","shell.execute_reply.started":"2022-06-25T21:31:27.624832Z","shell.execute_reply":"2022-06-25T21:31:27.769413Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['accuracy'])      # Accuracy learning curve of training set.\nplt.plot(history.history['val_accuracy'])  # Accuracy learning curve of validation set.\nplt.title('model accuracy')\nplt.ylabel('accuracy')\nplt.xlabel('epoch')\nplt.legend(['train', 'val'], loc='upper left')\nplt.title(\"Accuracy learning curves\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:27.771541Z","iopub.execute_input":"2022-06-25T21:31:27.772040Z","iopub.status.idle":"2022-06-25T21:31:27.921257Z","shell.execute_reply.started":"2022-06-25T21:31:27.772001Z","shell.execute_reply":"2022-06-25T21:31:27.920386Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### **confusion matrix**","metadata":{}},{"cell_type":"code","source":"y_predict=model.predict(x_val)\nconf_mat=tensorflow.math.confusion_matrix(np.argmax(y_val,axis=1) , np.argmax(y_predict,axis=1))","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:27.922521Z","iopub.execute_input":"2022-06-25T21:31:27.923009Z","iopub.status.idle":"2022-06-25T21:31:28.602041Z","shell.execute_reply.started":"2022-06-25T21:31:27.922970Z","shell.execute_reply":"2022-06-25T21:31:28.601136Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_cm = pd.DataFrame(np.array(conf_mat), index = [i for i in classes],\n                  columns = [i for i in classes])\nplt.figure(figsize = (13,7))\nax = sn.heatmap(df_cm, annot=True)\nplt.title(\"Confusion Matrix\", fontsize=20)\nplt.ylabel(\"True Class\"     , fontsize=20)\nplt.xlabel(\"Predicted Class\", fontsize=20)\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:28.604384Z","iopub.execute_input":"2022-06-25T21:31:28.604892Z","iopub.status.idle":"2022-06-25T21:31:29.209349Z","shell.execute_reply.started":"2022-06-25T21:31:28.604854Z","shell.execute_reply":"2022-06-25T21:31:29.208425Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### loading model to be tested locally.","metadata":{}},{"cell_type":"code","source":"from keras.models import load_model\nmodel.save(\"numbers_ANN.h5\")    ","metadata":{"execution":{"iopub.status.busy":"2022-06-25T21:31:29.210595Z","iopub.execute_input":"2022-06-25T21:31:29.211090Z","iopub.status.idle":"2022-06-25T21:31:29.255306Z","shell.execute_reply.started":"2022-06-25T21:31:29.211050Z","shell.execute_reply":"2022-06-25T21:31:29.254540Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}