diff --git a/TD2_Deep_Learning.ipynb b/TD2_Deep_Learning.ipynb index afff4d1c59351cc4335b705fa65916f1e0f98f8b..13294ff64f3b0f558293f18c84537757e1f5b3ee 100644 --- a/TD2_Deep_Learning.ipynb +++ b/TD2_Deep_Learning.ipynb @@ -692,6 +692,14 @@ ")" ] }, + { + "cell_type": "markdown", + "id": "df2b0014", + "metadata": {}, + "source": [ + "The accuracy is a lot better that the one of the neural network implemented in TD1 that was berely above 20%." + ] + }, { "cell_type": "markdown", "id": "944991a2", @@ -1062,6 +1070,14 @@ ")" ] }, + { + "cell_type": "markdown", + "id": "02200b5e", + "metadata": {}, + "source": [ + "The new model has a accuracy of 73% which is better that the previous model. It may be because of the dropout that allows the model to be more adapatable to new data." + ] + }, { "cell_type": "markdown", "id": "bc381cf4", @@ -1297,7 +1313,7 @@ "id": "c37a1007", "metadata": {}, "source": [ - "The compararison between the test accuracy of each class and the overall test accurcay shows that there is not a significant impact of the use of a quantized model on the prediction. The quantized model is \"only\" **0.07%** (nb of correct prediction of the original model-nb of correct prediction of the quantized model/nb of sample) less accurate than the original model but it is 2330.946/659.806 = **3,5** smaller." + "The compararison between the test accuracy of each class and the overall test accurcay shows that there is not a significant impact of the use of a quantized model on the prediction. The quantized model is only **0.07%** (nb of correct prediction of the original model-nb of correct prediction of the quantized model/nb of sample) less accurate than the original model but it is 2330.946/659.806 = **3,5** smaller !" ] }, { @@ -1770,7 +1786,7 @@ "id": "f435974d", "metadata": {}, "source": [ - "The quantized also predict the correct labels and is much faster than the original VGG model. It is also 3,03 (553439.178/182540.454) smaller.\n", + "The quantized also predict the correct labels and is much faster than the original VGG model. It is also **3,03** (553439.178/182540.454) smaller.\n", "\n", "Time of computation:\n", "- Original VGG: 6 min 21s\n",